]>
Commit | Line | Data |
---|---|---|
81819f0f CL |
1 | /* |
2 | * SLUB: A slab allocator that limits cache line use instead of queuing | |
3 | * objects in per cpu and per node lists. | |
4 | * | |
881db7fb CL |
5 | * The allocator synchronizes using per slab locks or atomic operatios |
6 | * and only uses a centralized lock to manage a pool of partial slabs. | |
81819f0f | 7 | * |
cde53535 | 8 | * (C) 2007 SGI, Christoph Lameter |
881db7fb | 9 | * (C) 2011 Linux Foundation, Christoph Lameter |
81819f0f CL |
10 | */ |
11 | ||
12 | #include <linux/mm.h> | |
1eb5ac64 | 13 | #include <linux/swap.h> /* struct reclaim_state */ |
81819f0f CL |
14 | #include <linux/module.h> |
15 | #include <linux/bit_spinlock.h> | |
16 | #include <linux/interrupt.h> | |
17 | #include <linux/bitops.h> | |
18 | #include <linux/slab.h> | |
97d06609 | 19 | #include "slab.h" |
7b3c3a50 | 20 | #include <linux/proc_fs.h> |
81819f0f | 21 | #include <linux/seq_file.h> |
5a896d9e | 22 | #include <linux/kmemcheck.h> |
81819f0f CL |
23 | #include <linux/cpu.h> |
24 | #include <linux/cpuset.h> | |
25 | #include <linux/mempolicy.h> | |
26 | #include <linux/ctype.h> | |
3ac7fe5a | 27 | #include <linux/debugobjects.h> |
81819f0f | 28 | #include <linux/kallsyms.h> |
b9049e23 | 29 | #include <linux/memory.h> |
f8bd2258 | 30 | #include <linux/math64.h> |
773ff60e | 31 | #include <linux/fault-inject.h> |
bfa71457 | 32 | #include <linux/stacktrace.h> |
4de900b4 | 33 | #include <linux/prefetch.h> |
81819f0f | 34 | |
4a92379b RK |
35 | #include <trace/events/kmem.h> |
36 | ||
81819f0f CL |
37 | /* |
38 | * Lock order: | |
881db7fb CL |
39 | * 1. slub_lock (Global Semaphore) |
40 | * 2. node->list_lock | |
41 | * 3. slab_lock(page) (Only on some arches and for debugging) | |
81819f0f | 42 | * |
881db7fb CL |
43 | * slub_lock |
44 | * | |
45 | * The role of the slub_lock is to protect the list of all the slabs | |
46 | * and to synchronize major metadata changes to slab cache structures. | |
47 | * | |
48 | * The slab_lock is only used for debugging and on arches that do not | |
49 | * have the ability to do a cmpxchg_double. It only protects the second | |
50 | * double word in the page struct. Meaning | |
51 | * A. page->freelist -> List of object free in a page | |
52 | * B. page->counters -> Counters of objects | |
53 | * C. page->frozen -> frozen state | |
54 | * | |
55 | * If a slab is frozen then it is exempt from list management. It is not | |
56 | * on any list. The processor that froze the slab is the one who can | |
57 | * perform list operations on the page. Other processors may put objects | |
58 | * onto the freelist but the processor that froze the slab is the only | |
59 | * one that can retrieve the objects from the page's freelist. | |
81819f0f CL |
60 | * |
61 | * The list_lock protects the partial and full list on each node and | |
62 | * the partial slab counter. If taken then no new slabs may be added or | |
63 | * removed from the lists nor make the number of partial slabs be modified. | |
64 | * (Note that the total number of slabs is an atomic value that may be | |
65 | * modified without taking the list lock). | |
66 | * | |
67 | * The list_lock is a centralized lock and thus we avoid taking it as | |
68 | * much as possible. As long as SLUB does not have to handle partial | |
69 | * slabs, operations can continue without any centralized lock. F.e. | |
70 | * allocating a long series of objects that fill up slabs does not require | |
71 | * the list lock. | |
81819f0f CL |
72 | * Interrupts are disabled during allocation and deallocation in order to |
73 | * make the slab allocator safe to use in the context of an irq. In addition | |
74 | * interrupts are disabled to ensure that the processor does not change | |
75 | * while handling per_cpu slabs, due to kernel preemption. | |
76 | * | |
77 | * SLUB assigns one slab for allocation to each processor. | |
78 | * Allocations only occur from these slabs called cpu slabs. | |
79 | * | |
672bba3a CL |
80 | * Slabs with free elements are kept on a partial list and during regular |
81 | * operations no list for full slabs is used. If an object in a full slab is | |
81819f0f | 82 | * freed then the slab will show up again on the partial lists. |
672bba3a CL |
83 | * We track full slabs for debugging purposes though because otherwise we |
84 | * cannot scan all objects. | |
81819f0f CL |
85 | * |
86 | * Slabs are freed when they become empty. Teardown and setup is | |
87 | * minimal so we rely on the page allocators per cpu caches for | |
88 | * fast frees and allocs. | |
89 | * | |
90 | * Overloading of page flags that are otherwise used for LRU management. | |
91 | * | |
4b6f0750 CL |
92 | * PageActive The slab is frozen and exempt from list processing. |
93 | * This means that the slab is dedicated to a purpose | |
94 | * such as satisfying allocations for a specific | |
95 | * processor. Objects may be freed in the slab while | |
96 | * it is frozen but slab_free will then skip the usual | |
97 | * list operations. It is up to the processor holding | |
98 | * the slab to integrate the slab into the slab lists | |
99 | * when the slab is no longer needed. | |
100 | * | |
101 | * One use of this flag is to mark slabs that are | |
102 | * used for allocations. Then such a slab becomes a cpu | |
103 | * slab. The cpu slab may be equipped with an additional | |
dfb4f096 | 104 | * freelist that allows lockless access to |
894b8788 CL |
105 | * free objects in addition to the regular freelist |
106 | * that requires the slab lock. | |
81819f0f CL |
107 | * |
108 | * PageError Slab requires special handling due to debug | |
109 | * options set. This moves slab handling out of | |
894b8788 | 110 | * the fast path and disables lockless freelists. |
81819f0f CL |
111 | */ |
112 | ||
af537b0a CL |
113 | #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ |
114 | SLAB_TRACE | SLAB_DEBUG_FREE) | |
115 | ||
116 | static inline int kmem_cache_debug(struct kmem_cache *s) | |
117 | { | |
5577bd8a | 118 | #ifdef CONFIG_SLUB_DEBUG |
af537b0a | 119 | return unlikely(s->flags & SLAB_DEBUG_FLAGS); |
5577bd8a | 120 | #else |
af537b0a | 121 | return 0; |
5577bd8a | 122 | #endif |
af537b0a | 123 | } |
5577bd8a | 124 | |
81819f0f CL |
125 | /* |
126 | * Issues still to be resolved: | |
127 | * | |
81819f0f CL |
128 | * - Support PAGE_ALLOC_DEBUG. Should be easy to do. |
129 | * | |
81819f0f CL |
130 | * - Variable sizing of the per node arrays |
131 | */ | |
132 | ||
133 | /* Enable to test recovery from slab corruption on boot */ | |
134 | #undef SLUB_RESILIENCY_TEST | |
135 | ||
b789ef51 CL |
136 | /* Enable to log cmpxchg failures */ |
137 | #undef SLUB_DEBUG_CMPXCHG | |
138 | ||
2086d26a CL |
139 | /* |
140 | * Mininum number of partial slabs. These will be left on the partial | |
141 | * lists even if they are empty. kmem_cache_shrink may reclaim them. | |
142 | */ | |
76be8950 | 143 | #define MIN_PARTIAL 5 |
e95eed57 | 144 | |
2086d26a CL |
145 | /* |
146 | * Maximum number of desirable partial slabs. | |
147 | * The existence of more partial slabs makes kmem_cache_shrink | |
148 | * sort the partial list by the number of objects in the. | |
149 | */ | |
150 | #define MAX_PARTIAL 10 | |
151 | ||
81819f0f CL |
152 | #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \ |
153 | SLAB_POISON | SLAB_STORE_USER) | |
672bba3a | 154 | |
fa5ec8a1 | 155 | /* |
3de47213 DR |
156 | * Debugging flags that require metadata to be stored in the slab. These get |
157 | * disabled when slub_debug=O is used and a cache's min order increases with | |
158 | * metadata. | |
fa5ec8a1 | 159 | */ |
3de47213 | 160 | #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER) |
fa5ec8a1 | 161 | |
81819f0f CL |
162 | /* |
163 | * Set of flags that will prevent slab merging | |
164 | */ | |
165 | #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ | |
4c13dd3b DM |
166 | SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \ |
167 | SLAB_FAILSLAB) | |
81819f0f CL |
168 | |
169 | #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \ | |
5a896d9e | 170 | SLAB_CACHE_DMA | SLAB_NOTRACK) |
81819f0f | 171 | |
210b5c06 CG |
172 | #define OO_SHIFT 16 |
173 | #define OO_MASK ((1 << OO_SHIFT) - 1) | |
50d5c41c | 174 | #define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */ |
210b5c06 | 175 | |
81819f0f | 176 | /* Internal SLUB flags */ |
f90ec390 | 177 | #define __OBJECT_POISON 0x80000000UL /* Poison object */ |
b789ef51 | 178 | #define __CMPXCHG_DOUBLE 0x40000000UL /* Use cmpxchg_double */ |
81819f0f CL |
179 | |
180 | static int kmem_size = sizeof(struct kmem_cache); | |
181 | ||
182 | #ifdef CONFIG_SMP | |
183 | static struct notifier_block slab_notifier; | |
184 | #endif | |
185 | ||
81819f0f CL |
186 | /* A list of all slab caches on the system */ |
187 | static DECLARE_RWSEM(slub_lock); | |
5af328a5 | 188 | static LIST_HEAD(slab_caches); |
81819f0f | 189 | |
02cbc874 CL |
190 | /* |
191 | * Tracking user of a slab. | |
192 | */ | |
d6543e39 | 193 | #define TRACK_ADDRS_COUNT 16 |
02cbc874 | 194 | struct track { |
ce71e27c | 195 | unsigned long addr; /* Called from address */ |
d6543e39 BG |
196 | #ifdef CONFIG_STACKTRACE |
197 | unsigned long addrs[TRACK_ADDRS_COUNT]; /* Called from address */ | |
198 | #endif | |
02cbc874 CL |
199 | int cpu; /* Was running on cpu */ |
200 | int pid; /* Pid context */ | |
201 | unsigned long when; /* When did the operation occur */ | |
202 | }; | |
203 | ||
204 | enum track_item { TRACK_ALLOC, TRACK_FREE }; | |
205 | ||
ab4d5ed5 | 206 | #ifdef CONFIG_SYSFS |
81819f0f CL |
207 | static int sysfs_slab_add(struct kmem_cache *); |
208 | static int sysfs_slab_alias(struct kmem_cache *, const char *); | |
209 | static void sysfs_slab_remove(struct kmem_cache *); | |
8ff12cfc | 210 | |
81819f0f | 211 | #else |
0c710013 CL |
212 | static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; } |
213 | static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p) | |
214 | { return 0; } | |
151c602f CL |
215 | static inline void sysfs_slab_remove(struct kmem_cache *s) |
216 | { | |
84c1cf62 | 217 | kfree(s->name); |
151c602f CL |
218 | kfree(s); |
219 | } | |
8ff12cfc | 220 | |
81819f0f CL |
221 | #endif |
222 | ||
4fdccdfb | 223 | static inline void stat(const struct kmem_cache *s, enum stat_item si) |
8ff12cfc CL |
224 | { |
225 | #ifdef CONFIG_SLUB_STATS | |
84e554e6 | 226 | __this_cpu_inc(s->cpu_slab->stat[si]); |
8ff12cfc CL |
227 | #endif |
228 | } | |
229 | ||
81819f0f CL |
230 | /******************************************************************** |
231 | * Core slab cache functions | |
232 | *******************************************************************/ | |
233 | ||
81819f0f CL |
234 | static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node) |
235 | { | |
81819f0f | 236 | return s->node[node]; |
81819f0f CL |
237 | } |
238 | ||
6446faa2 | 239 | /* Verify that a pointer has an address that is valid within a slab page */ |
02cbc874 CL |
240 | static inline int check_valid_pointer(struct kmem_cache *s, |
241 | struct page *page, const void *object) | |
242 | { | |
243 | void *base; | |
244 | ||
a973e9dd | 245 | if (!object) |
02cbc874 CL |
246 | return 1; |
247 | ||
a973e9dd | 248 | base = page_address(page); |
39b26464 | 249 | if (object < base || object >= base + page->objects * s->size || |
02cbc874 CL |
250 | (object - base) % s->size) { |
251 | return 0; | |
252 | } | |
253 | ||
254 | return 1; | |
255 | } | |
256 | ||
7656c72b CL |
257 | static inline void *get_freepointer(struct kmem_cache *s, void *object) |
258 | { | |
259 | return *(void **)(object + s->offset); | |
260 | } | |
261 | ||
0ad9500e ED |
262 | static void prefetch_freepointer(const struct kmem_cache *s, void *object) |
263 | { | |
264 | prefetch(object + s->offset); | |
265 | } | |
266 | ||
1393d9a1 CL |
267 | static inline void *get_freepointer_safe(struct kmem_cache *s, void *object) |
268 | { | |
269 | void *p; | |
270 | ||
271 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
272 | probe_kernel_read(&p, (void **)(object + s->offset), sizeof(p)); | |
273 | #else | |
274 | p = get_freepointer(s, object); | |
275 | #endif | |
276 | return p; | |
277 | } | |
278 | ||
7656c72b CL |
279 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) |
280 | { | |
281 | *(void **)(object + s->offset) = fp; | |
282 | } | |
283 | ||
284 | /* Loop over all objects in a slab */ | |
224a88be CL |
285 | #define for_each_object(__p, __s, __addr, __objects) \ |
286 | for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\ | |
7656c72b CL |
287 | __p += (__s)->size) |
288 | ||
7656c72b CL |
289 | /* Determine object index from a given position */ |
290 | static inline int slab_index(void *p, struct kmem_cache *s, void *addr) | |
291 | { | |
292 | return (p - addr) / s->size; | |
293 | } | |
294 | ||
d71f606f MK |
295 | static inline size_t slab_ksize(const struct kmem_cache *s) |
296 | { | |
297 | #ifdef CONFIG_SLUB_DEBUG | |
298 | /* | |
299 | * Debugging requires use of the padding between object | |
300 | * and whatever may come after it. | |
301 | */ | |
302 | if (s->flags & (SLAB_RED_ZONE | SLAB_POISON)) | |
3b0efdfa | 303 | return s->object_size; |
d71f606f MK |
304 | |
305 | #endif | |
306 | /* | |
307 | * If we have the need to store the freelist pointer | |
308 | * back there or track user information then we can | |
309 | * only use the space before that information. | |
310 | */ | |
311 | if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER)) | |
312 | return s->inuse; | |
313 | /* | |
314 | * Else we can use all the padding etc for the allocation | |
315 | */ | |
316 | return s->size; | |
317 | } | |
318 | ||
ab9a0f19 LJ |
319 | static inline int order_objects(int order, unsigned long size, int reserved) |
320 | { | |
321 | return ((PAGE_SIZE << order) - reserved) / size; | |
322 | } | |
323 | ||
834f3d11 | 324 | static inline struct kmem_cache_order_objects oo_make(int order, |
ab9a0f19 | 325 | unsigned long size, int reserved) |
834f3d11 CL |
326 | { |
327 | struct kmem_cache_order_objects x = { | |
ab9a0f19 | 328 | (order << OO_SHIFT) + order_objects(order, size, reserved) |
834f3d11 CL |
329 | }; |
330 | ||
331 | return x; | |
332 | } | |
333 | ||
334 | static inline int oo_order(struct kmem_cache_order_objects x) | |
335 | { | |
210b5c06 | 336 | return x.x >> OO_SHIFT; |
834f3d11 CL |
337 | } |
338 | ||
339 | static inline int oo_objects(struct kmem_cache_order_objects x) | |
340 | { | |
210b5c06 | 341 | return x.x & OO_MASK; |
834f3d11 CL |
342 | } |
343 | ||
881db7fb CL |
344 | /* |
345 | * Per slab locking using the pagelock | |
346 | */ | |
347 | static __always_inline void slab_lock(struct page *page) | |
348 | { | |
349 | bit_spin_lock(PG_locked, &page->flags); | |
350 | } | |
351 | ||
352 | static __always_inline void slab_unlock(struct page *page) | |
353 | { | |
354 | __bit_spin_unlock(PG_locked, &page->flags); | |
355 | } | |
356 | ||
1d07171c CL |
357 | /* Interrupts must be disabled (for the fallback code to work right) */ |
358 | static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct page *page, | |
359 | void *freelist_old, unsigned long counters_old, | |
360 | void *freelist_new, unsigned long counters_new, | |
361 | const char *n) | |
362 | { | |
363 | VM_BUG_ON(!irqs_disabled()); | |
2565409f HC |
364 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
365 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
1d07171c | 366 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 367 | if (cmpxchg_double(&page->freelist, &page->counters, |
1d07171c CL |
368 | freelist_old, counters_old, |
369 | freelist_new, counters_new)) | |
370 | return 1; | |
371 | } else | |
372 | #endif | |
373 | { | |
374 | slab_lock(page); | |
375 | if (page->freelist == freelist_old && page->counters == counters_old) { | |
376 | page->freelist = freelist_new; | |
377 | page->counters = counters_new; | |
378 | slab_unlock(page); | |
379 | return 1; | |
380 | } | |
381 | slab_unlock(page); | |
382 | } | |
383 | ||
384 | cpu_relax(); | |
385 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
386 | ||
387 | #ifdef SLUB_DEBUG_CMPXCHG | |
388 | printk(KERN_INFO "%s %s: cmpxchg double redo ", n, s->name); | |
389 | #endif | |
390 | ||
391 | return 0; | |
392 | } | |
393 | ||
b789ef51 CL |
394 | static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page, |
395 | void *freelist_old, unsigned long counters_old, | |
396 | void *freelist_new, unsigned long counters_new, | |
397 | const char *n) | |
398 | { | |
2565409f HC |
399 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
400 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
b789ef51 | 401 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 402 | if (cmpxchg_double(&page->freelist, &page->counters, |
b789ef51 CL |
403 | freelist_old, counters_old, |
404 | freelist_new, counters_new)) | |
405 | return 1; | |
406 | } else | |
407 | #endif | |
408 | { | |
1d07171c CL |
409 | unsigned long flags; |
410 | ||
411 | local_irq_save(flags); | |
881db7fb | 412 | slab_lock(page); |
b789ef51 CL |
413 | if (page->freelist == freelist_old && page->counters == counters_old) { |
414 | page->freelist = freelist_new; | |
415 | page->counters = counters_new; | |
881db7fb | 416 | slab_unlock(page); |
1d07171c | 417 | local_irq_restore(flags); |
b789ef51 CL |
418 | return 1; |
419 | } | |
881db7fb | 420 | slab_unlock(page); |
1d07171c | 421 | local_irq_restore(flags); |
b789ef51 CL |
422 | } |
423 | ||
424 | cpu_relax(); | |
425 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
426 | ||
427 | #ifdef SLUB_DEBUG_CMPXCHG | |
428 | printk(KERN_INFO "%s %s: cmpxchg double redo ", n, s->name); | |
429 | #endif | |
430 | ||
431 | return 0; | |
432 | } | |
433 | ||
41ecc55b | 434 | #ifdef CONFIG_SLUB_DEBUG |
5f80b13a CL |
435 | /* |
436 | * Determine a map of object in use on a page. | |
437 | * | |
881db7fb | 438 | * Node listlock must be held to guarantee that the page does |
5f80b13a CL |
439 | * not vanish from under us. |
440 | */ | |
441 | static void get_map(struct kmem_cache *s, struct page *page, unsigned long *map) | |
442 | { | |
443 | void *p; | |
444 | void *addr = page_address(page); | |
445 | ||
446 | for (p = page->freelist; p; p = get_freepointer(s, p)) | |
447 | set_bit(slab_index(p, s, addr), map); | |
448 | } | |
449 | ||
41ecc55b CL |
450 | /* |
451 | * Debug settings: | |
452 | */ | |
f0630fff CL |
453 | #ifdef CONFIG_SLUB_DEBUG_ON |
454 | static int slub_debug = DEBUG_DEFAULT_FLAGS; | |
455 | #else | |
41ecc55b | 456 | static int slub_debug; |
f0630fff | 457 | #endif |
41ecc55b CL |
458 | |
459 | static char *slub_debug_slabs; | |
fa5ec8a1 | 460 | static int disable_higher_order_debug; |
41ecc55b | 461 | |
81819f0f CL |
462 | /* |
463 | * Object debugging | |
464 | */ | |
465 | static void print_section(char *text, u8 *addr, unsigned int length) | |
466 | { | |
ffc79d28 SAS |
467 | print_hex_dump(KERN_ERR, text, DUMP_PREFIX_ADDRESS, 16, 1, addr, |
468 | length, 1); | |
81819f0f CL |
469 | } |
470 | ||
81819f0f CL |
471 | static struct track *get_track(struct kmem_cache *s, void *object, |
472 | enum track_item alloc) | |
473 | { | |
474 | struct track *p; | |
475 | ||
476 | if (s->offset) | |
477 | p = object + s->offset + sizeof(void *); | |
478 | else | |
479 | p = object + s->inuse; | |
480 | ||
481 | return p + alloc; | |
482 | } | |
483 | ||
484 | static void set_track(struct kmem_cache *s, void *object, | |
ce71e27c | 485 | enum track_item alloc, unsigned long addr) |
81819f0f | 486 | { |
1a00df4a | 487 | struct track *p = get_track(s, object, alloc); |
81819f0f | 488 | |
81819f0f | 489 | if (addr) { |
d6543e39 BG |
490 | #ifdef CONFIG_STACKTRACE |
491 | struct stack_trace trace; | |
492 | int i; | |
493 | ||
494 | trace.nr_entries = 0; | |
495 | trace.max_entries = TRACK_ADDRS_COUNT; | |
496 | trace.entries = p->addrs; | |
497 | trace.skip = 3; | |
498 | save_stack_trace(&trace); | |
499 | ||
500 | /* See rant in lockdep.c */ | |
501 | if (trace.nr_entries != 0 && | |
502 | trace.entries[trace.nr_entries - 1] == ULONG_MAX) | |
503 | trace.nr_entries--; | |
504 | ||
505 | for (i = trace.nr_entries; i < TRACK_ADDRS_COUNT; i++) | |
506 | p->addrs[i] = 0; | |
507 | #endif | |
81819f0f CL |
508 | p->addr = addr; |
509 | p->cpu = smp_processor_id(); | |
88e4ccf2 | 510 | p->pid = current->pid; |
81819f0f CL |
511 | p->when = jiffies; |
512 | } else | |
513 | memset(p, 0, sizeof(struct track)); | |
514 | } | |
515 | ||
81819f0f CL |
516 | static void init_tracking(struct kmem_cache *s, void *object) |
517 | { | |
24922684 CL |
518 | if (!(s->flags & SLAB_STORE_USER)) |
519 | return; | |
520 | ||
ce71e27c EGM |
521 | set_track(s, object, TRACK_FREE, 0UL); |
522 | set_track(s, object, TRACK_ALLOC, 0UL); | |
81819f0f CL |
523 | } |
524 | ||
525 | static void print_track(const char *s, struct track *t) | |
526 | { | |
527 | if (!t->addr) | |
528 | return; | |
529 | ||
7daf705f | 530 | printk(KERN_ERR "INFO: %s in %pS age=%lu cpu=%u pid=%d\n", |
ce71e27c | 531 | s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid); |
d6543e39 BG |
532 | #ifdef CONFIG_STACKTRACE |
533 | { | |
534 | int i; | |
535 | for (i = 0; i < TRACK_ADDRS_COUNT; i++) | |
536 | if (t->addrs[i]) | |
537 | printk(KERN_ERR "\t%pS\n", (void *)t->addrs[i]); | |
538 | else | |
539 | break; | |
540 | } | |
541 | #endif | |
24922684 CL |
542 | } |
543 | ||
544 | static void print_tracking(struct kmem_cache *s, void *object) | |
545 | { | |
546 | if (!(s->flags & SLAB_STORE_USER)) | |
547 | return; | |
548 | ||
549 | print_track("Allocated", get_track(s, object, TRACK_ALLOC)); | |
550 | print_track("Freed", get_track(s, object, TRACK_FREE)); | |
551 | } | |
552 | ||
553 | static void print_page_info(struct page *page) | |
554 | { | |
39b26464 CL |
555 | printk(KERN_ERR "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n", |
556 | page, page->objects, page->inuse, page->freelist, page->flags); | |
24922684 CL |
557 | |
558 | } | |
559 | ||
560 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
561 | { | |
562 | va_list args; | |
563 | char buf[100]; | |
564 | ||
565 | va_start(args, fmt); | |
566 | vsnprintf(buf, sizeof(buf), fmt, args); | |
567 | va_end(args); | |
568 | printk(KERN_ERR "========================================" | |
569 | "=====================================\n"); | |
265d47e7 | 570 | printk(KERN_ERR "BUG %s (%s): %s\n", s->name, print_tainted(), buf); |
24922684 CL |
571 | printk(KERN_ERR "----------------------------------------" |
572 | "-------------------------------------\n\n"); | |
81819f0f CL |
573 | } |
574 | ||
24922684 CL |
575 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
576 | { | |
577 | va_list args; | |
578 | char buf[100]; | |
579 | ||
580 | va_start(args, fmt); | |
581 | vsnprintf(buf, sizeof(buf), fmt, args); | |
582 | va_end(args); | |
583 | printk(KERN_ERR "FIX %s: %s\n", s->name, buf); | |
584 | } | |
585 | ||
586 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
81819f0f CL |
587 | { |
588 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 589 | u8 *addr = page_address(page); |
24922684 CL |
590 | |
591 | print_tracking(s, p); | |
592 | ||
593 | print_page_info(page); | |
594 | ||
595 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", | |
596 | p, p - addr, get_freepointer(s, p)); | |
597 | ||
598 | if (p > addr + 16) | |
ffc79d28 | 599 | print_section("Bytes b4 ", p - 16, 16); |
81819f0f | 600 | |
3b0efdfa | 601 | print_section("Object ", p, min_t(unsigned long, s->object_size, |
ffc79d28 | 602 | PAGE_SIZE)); |
81819f0f | 603 | if (s->flags & SLAB_RED_ZONE) |
3b0efdfa CL |
604 | print_section("Redzone ", p + s->object_size, |
605 | s->inuse - s->object_size); | |
81819f0f | 606 | |
81819f0f CL |
607 | if (s->offset) |
608 | off = s->offset + sizeof(void *); | |
609 | else | |
610 | off = s->inuse; | |
611 | ||
24922684 | 612 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 613 | off += 2 * sizeof(struct track); |
81819f0f CL |
614 | |
615 | if (off != s->size) | |
616 | /* Beginning of the filler is the free pointer */ | |
ffc79d28 | 617 | print_section("Padding ", p + off, s->size - off); |
24922684 CL |
618 | |
619 | dump_stack(); | |
81819f0f CL |
620 | } |
621 | ||
622 | static void object_err(struct kmem_cache *s, struct page *page, | |
623 | u8 *object, char *reason) | |
624 | { | |
3dc50637 | 625 | slab_bug(s, "%s", reason); |
24922684 | 626 | print_trailer(s, page, object); |
81819f0f CL |
627 | } |
628 | ||
24922684 | 629 | static void slab_err(struct kmem_cache *s, struct page *page, char *fmt, ...) |
81819f0f CL |
630 | { |
631 | va_list args; | |
632 | char buf[100]; | |
633 | ||
24922684 CL |
634 | va_start(args, fmt); |
635 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 636 | va_end(args); |
3dc50637 | 637 | slab_bug(s, "%s", buf); |
24922684 | 638 | print_page_info(page); |
81819f0f CL |
639 | dump_stack(); |
640 | } | |
641 | ||
f7cb1933 | 642 | static void init_object(struct kmem_cache *s, void *object, u8 val) |
81819f0f CL |
643 | { |
644 | u8 *p = object; | |
645 | ||
646 | if (s->flags & __OBJECT_POISON) { | |
3b0efdfa CL |
647 | memset(p, POISON_FREE, s->object_size - 1); |
648 | p[s->object_size - 1] = POISON_END; | |
81819f0f CL |
649 | } |
650 | ||
651 | if (s->flags & SLAB_RED_ZONE) | |
3b0efdfa | 652 | memset(p + s->object_size, val, s->inuse - s->object_size); |
81819f0f CL |
653 | } |
654 | ||
24922684 CL |
655 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, |
656 | void *from, void *to) | |
657 | { | |
658 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
659 | memset(from, data, to - from); | |
660 | } | |
661 | ||
662 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
663 | u8 *object, char *what, | |
06428780 | 664 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
665 | { |
666 | u8 *fault; | |
667 | u8 *end; | |
668 | ||
79824820 | 669 | fault = memchr_inv(start, value, bytes); |
24922684 CL |
670 | if (!fault) |
671 | return 1; | |
672 | ||
673 | end = start + bytes; | |
674 | while (end > fault && end[-1] == value) | |
675 | end--; | |
676 | ||
677 | slab_bug(s, "%s overwritten", what); | |
678 | printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n", | |
679 | fault, end - 1, fault[0], value); | |
680 | print_trailer(s, page, object); | |
681 | ||
682 | restore_bytes(s, what, value, fault, end); | |
683 | return 0; | |
81819f0f CL |
684 | } |
685 | ||
81819f0f CL |
686 | /* |
687 | * Object layout: | |
688 | * | |
689 | * object address | |
690 | * Bytes of the object to be managed. | |
691 | * If the freepointer may overlay the object then the free | |
692 | * pointer is the first word of the object. | |
672bba3a | 693 | * |
81819f0f CL |
694 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
695 | * 0xa5 (POISON_END) | |
696 | * | |
3b0efdfa | 697 | * object + s->object_size |
81819f0f | 698 | * Padding to reach word boundary. This is also used for Redzoning. |
672bba3a | 699 | * Padding is extended by another word if Redzoning is enabled and |
3b0efdfa | 700 | * object_size == inuse. |
672bba3a | 701 | * |
81819f0f CL |
702 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
703 | * 0xcc (RED_ACTIVE) for objects in use. | |
704 | * | |
705 | * object + s->inuse | |
672bba3a CL |
706 | * Meta data starts here. |
707 | * | |
81819f0f CL |
708 | * A. Free pointer (if we cannot overwrite object on free) |
709 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a | 710 | * C. Padding to reach required alignment boundary or at mininum |
6446faa2 | 711 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
712 | * before the word boundary. |
713 | * | |
714 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
715 | * |
716 | * object + s->size | |
672bba3a | 717 | * Nothing is used beyond s->size. |
81819f0f | 718 | * |
3b0efdfa | 719 | * If slabcaches are merged then the object_size and inuse boundaries are mostly |
672bba3a | 720 | * ignored. And therefore no slab options that rely on these boundaries |
81819f0f CL |
721 | * may be used with merged slabcaches. |
722 | */ | |
723 | ||
81819f0f CL |
724 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
725 | { | |
726 | unsigned long off = s->inuse; /* The end of info */ | |
727 | ||
728 | if (s->offset) | |
729 | /* Freepointer is placed after the object. */ | |
730 | off += sizeof(void *); | |
731 | ||
732 | if (s->flags & SLAB_STORE_USER) | |
733 | /* We also have user information there */ | |
734 | off += 2 * sizeof(struct track); | |
735 | ||
736 | if (s->size == off) | |
737 | return 1; | |
738 | ||
24922684 CL |
739 | return check_bytes_and_report(s, page, p, "Object padding", |
740 | p + off, POISON_INUSE, s->size - off); | |
81819f0f CL |
741 | } |
742 | ||
39b26464 | 743 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
744 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
745 | { | |
24922684 CL |
746 | u8 *start; |
747 | u8 *fault; | |
748 | u8 *end; | |
749 | int length; | |
750 | int remainder; | |
81819f0f CL |
751 | |
752 | if (!(s->flags & SLAB_POISON)) | |
753 | return 1; | |
754 | ||
a973e9dd | 755 | start = page_address(page); |
ab9a0f19 | 756 | length = (PAGE_SIZE << compound_order(page)) - s->reserved; |
39b26464 CL |
757 | end = start + length; |
758 | remainder = length % s->size; | |
81819f0f CL |
759 | if (!remainder) |
760 | return 1; | |
761 | ||
79824820 | 762 | fault = memchr_inv(end - remainder, POISON_INUSE, remainder); |
24922684 CL |
763 | if (!fault) |
764 | return 1; | |
765 | while (end > fault && end[-1] == POISON_INUSE) | |
766 | end--; | |
767 | ||
768 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1); | |
ffc79d28 | 769 | print_section("Padding ", end - remainder, remainder); |
24922684 | 770 | |
8a3d271d | 771 | restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end); |
24922684 | 772 | return 0; |
81819f0f CL |
773 | } |
774 | ||
775 | static int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 776 | void *object, u8 val) |
81819f0f CL |
777 | { |
778 | u8 *p = object; | |
3b0efdfa | 779 | u8 *endobject = object + s->object_size; |
81819f0f CL |
780 | |
781 | if (s->flags & SLAB_RED_ZONE) { | |
24922684 | 782 | if (!check_bytes_and_report(s, page, object, "Redzone", |
3b0efdfa | 783 | endobject, val, s->inuse - s->object_size)) |
81819f0f | 784 | return 0; |
81819f0f | 785 | } else { |
3b0efdfa | 786 | if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) { |
3adbefee | 787 | check_bytes_and_report(s, page, p, "Alignment padding", |
3b0efdfa | 788 | endobject, POISON_INUSE, s->inuse - s->object_size); |
3adbefee | 789 | } |
81819f0f CL |
790 | } |
791 | ||
792 | if (s->flags & SLAB_POISON) { | |
f7cb1933 | 793 | if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) && |
24922684 | 794 | (!check_bytes_and_report(s, page, p, "Poison", p, |
3b0efdfa | 795 | POISON_FREE, s->object_size - 1) || |
24922684 | 796 | !check_bytes_and_report(s, page, p, "Poison", |
3b0efdfa | 797 | p + s->object_size - 1, POISON_END, 1))) |
81819f0f | 798 | return 0; |
81819f0f CL |
799 | /* |
800 | * check_pad_bytes cleans up on its own. | |
801 | */ | |
802 | check_pad_bytes(s, page, p); | |
803 | } | |
804 | ||
f7cb1933 | 805 | if (!s->offset && val == SLUB_RED_ACTIVE) |
81819f0f CL |
806 | /* |
807 | * Object and freepointer overlap. Cannot check | |
808 | * freepointer while object is allocated. | |
809 | */ | |
810 | return 1; | |
811 | ||
812 | /* Check free pointer validity */ | |
813 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
814 | object_err(s, page, p, "Freepointer corrupt"); | |
815 | /* | |
9f6c708e | 816 | * No choice but to zap it and thus lose the remainder |
81819f0f | 817 | * of the free objects in this slab. May cause |
672bba3a | 818 | * another error because the object count is now wrong. |
81819f0f | 819 | */ |
a973e9dd | 820 | set_freepointer(s, p, NULL); |
81819f0f CL |
821 | return 0; |
822 | } | |
823 | return 1; | |
824 | } | |
825 | ||
826 | static int check_slab(struct kmem_cache *s, struct page *page) | |
827 | { | |
39b26464 CL |
828 | int maxobj; |
829 | ||
81819f0f CL |
830 | VM_BUG_ON(!irqs_disabled()); |
831 | ||
832 | if (!PageSlab(page)) { | |
24922684 | 833 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
834 | return 0; |
835 | } | |
39b26464 | 836 | |
ab9a0f19 | 837 | maxobj = order_objects(compound_order(page), s->size, s->reserved); |
39b26464 CL |
838 | if (page->objects > maxobj) { |
839 | slab_err(s, page, "objects %u > max %u", | |
840 | s->name, page->objects, maxobj); | |
841 | return 0; | |
842 | } | |
843 | if (page->inuse > page->objects) { | |
24922684 | 844 | slab_err(s, page, "inuse %u > max %u", |
39b26464 | 845 | s->name, page->inuse, page->objects); |
81819f0f CL |
846 | return 0; |
847 | } | |
848 | /* Slab_pad_check fixes things up after itself */ | |
849 | slab_pad_check(s, page); | |
850 | return 1; | |
851 | } | |
852 | ||
853 | /* | |
672bba3a CL |
854 | * Determine if a certain object on a page is on the freelist. Must hold the |
855 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
856 | */ |
857 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
858 | { | |
859 | int nr = 0; | |
881db7fb | 860 | void *fp; |
81819f0f | 861 | void *object = NULL; |
224a88be | 862 | unsigned long max_objects; |
81819f0f | 863 | |
881db7fb | 864 | fp = page->freelist; |
39b26464 | 865 | while (fp && nr <= page->objects) { |
81819f0f CL |
866 | if (fp == search) |
867 | return 1; | |
868 | if (!check_valid_pointer(s, page, fp)) { | |
869 | if (object) { | |
870 | object_err(s, page, object, | |
871 | "Freechain corrupt"); | |
a973e9dd | 872 | set_freepointer(s, object, NULL); |
81819f0f CL |
873 | break; |
874 | } else { | |
24922684 | 875 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 876 | page->freelist = NULL; |
39b26464 | 877 | page->inuse = page->objects; |
24922684 | 878 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
879 | return 0; |
880 | } | |
881 | break; | |
882 | } | |
883 | object = fp; | |
884 | fp = get_freepointer(s, object); | |
885 | nr++; | |
886 | } | |
887 | ||
ab9a0f19 | 888 | max_objects = order_objects(compound_order(page), s->size, s->reserved); |
210b5c06 CG |
889 | if (max_objects > MAX_OBJS_PER_PAGE) |
890 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be CL |
891 | |
892 | if (page->objects != max_objects) { | |
893 | slab_err(s, page, "Wrong number of objects. Found %d but " | |
894 | "should be %d", page->objects, max_objects); | |
895 | page->objects = max_objects; | |
896 | slab_fix(s, "Number of objects adjusted."); | |
897 | } | |
39b26464 | 898 | if (page->inuse != page->objects - nr) { |
70d71228 | 899 | slab_err(s, page, "Wrong object count. Counter is %d but " |
39b26464 CL |
900 | "counted were %d", page->inuse, page->objects - nr); |
901 | page->inuse = page->objects - nr; | |
24922684 | 902 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
903 | } |
904 | return search == NULL; | |
905 | } | |
906 | ||
0121c619 CL |
907 | static void trace(struct kmem_cache *s, struct page *page, void *object, |
908 | int alloc) | |
3ec09742 CL |
909 | { |
910 | if (s->flags & SLAB_TRACE) { | |
911 | printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n", | |
912 | s->name, | |
913 | alloc ? "alloc" : "free", | |
914 | object, page->inuse, | |
915 | page->freelist); | |
916 | ||
917 | if (!alloc) | |
3b0efdfa | 918 | print_section("Object ", (void *)object, s->object_size); |
3ec09742 CL |
919 | |
920 | dump_stack(); | |
921 | } | |
922 | } | |
923 | ||
c016b0bd CL |
924 | /* |
925 | * Hooks for other subsystems that check memory allocations. In a typical | |
926 | * production configuration these hooks all should produce no code at all. | |
927 | */ | |
928 | static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags) | |
929 | { | |
c1d50836 | 930 | flags &= gfp_allowed_mask; |
c016b0bd CL |
931 | lockdep_trace_alloc(flags); |
932 | might_sleep_if(flags & __GFP_WAIT); | |
933 | ||
3b0efdfa | 934 | return should_failslab(s->object_size, flags, s->flags); |
c016b0bd CL |
935 | } |
936 | ||
937 | static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, void *object) | |
938 | { | |
c1d50836 | 939 | flags &= gfp_allowed_mask; |
b3d41885 | 940 | kmemcheck_slab_alloc(s, flags, object, slab_ksize(s)); |
3b0efdfa | 941 | kmemleak_alloc_recursive(object, s->object_size, 1, s->flags, flags); |
c016b0bd CL |
942 | } |
943 | ||
944 | static inline void slab_free_hook(struct kmem_cache *s, void *x) | |
945 | { | |
946 | kmemleak_free_recursive(x, s->flags); | |
c016b0bd | 947 | |
d3f661d6 CL |
948 | /* |
949 | * Trouble is that we may no longer disable interupts in the fast path | |
950 | * So in order to make the debug calls that expect irqs to be | |
951 | * disabled we need to disable interrupts temporarily. | |
952 | */ | |
953 | #if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP) | |
954 | { | |
955 | unsigned long flags; | |
956 | ||
957 | local_irq_save(flags); | |
3b0efdfa CL |
958 | kmemcheck_slab_free(s, x, s->object_size); |
959 | debug_check_no_locks_freed(x, s->object_size); | |
d3f661d6 CL |
960 | local_irq_restore(flags); |
961 | } | |
962 | #endif | |
f9b615de | 963 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) |
3b0efdfa | 964 | debug_check_no_obj_freed(x, s->object_size); |
c016b0bd CL |
965 | } |
966 | ||
643b1138 | 967 | /* |
672bba3a | 968 | * Tracking of fully allocated slabs for debugging purposes. |
5cc6eee8 CL |
969 | * |
970 | * list_lock must be held. | |
643b1138 | 971 | */ |
5cc6eee8 CL |
972 | static void add_full(struct kmem_cache *s, |
973 | struct kmem_cache_node *n, struct page *page) | |
643b1138 | 974 | { |
5cc6eee8 CL |
975 | if (!(s->flags & SLAB_STORE_USER)) |
976 | return; | |
977 | ||
643b1138 | 978 | list_add(&page->lru, &n->full); |
643b1138 CL |
979 | } |
980 | ||
5cc6eee8 CL |
981 | /* |
982 | * list_lock must be held. | |
983 | */ | |
643b1138 CL |
984 | static void remove_full(struct kmem_cache *s, struct page *page) |
985 | { | |
643b1138 CL |
986 | if (!(s->flags & SLAB_STORE_USER)) |
987 | return; | |
988 | ||
643b1138 | 989 | list_del(&page->lru); |
643b1138 CL |
990 | } |
991 | ||
0f389ec6 CL |
992 | /* Tracking of the number of slabs for debugging purposes */ |
993 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
994 | { | |
995 | struct kmem_cache_node *n = get_node(s, node); | |
996 | ||
997 | return atomic_long_read(&n->nr_slabs); | |
998 | } | |
999 | ||
26c02cf0 AB |
1000 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1001 | { | |
1002 | return atomic_long_read(&n->nr_slabs); | |
1003 | } | |
1004 | ||
205ab99d | 1005 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1006 | { |
1007 | struct kmem_cache_node *n = get_node(s, node); | |
1008 | ||
1009 | /* | |
1010 | * May be called early in order to allocate a slab for the | |
1011 | * kmem_cache_node structure. Solve the chicken-egg | |
1012 | * dilemma by deferring the increment of the count during | |
1013 | * bootstrap (see early_kmem_cache_node_alloc). | |
1014 | */ | |
7340cc84 | 1015 | if (n) { |
0f389ec6 | 1016 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
1017 | atomic_long_add(objects, &n->total_objects); |
1018 | } | |
0f389ec6 | 1019 | } |
205ab99d | 1020 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1021 | { |
1022 | struct kmem_cache_node *n = get_node(s, node); | |
1023 | ||
1024 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 1025 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
1026 | } |
1027 | ||
1028 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
1029 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
1030 | void *object) | |
1031 | { | |
1032 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
1033 | return; | |
1034 | ||
f7cb1933 | 1035 | init_object(s, object, SLUB_RED_INACTIVE); |
3ec09742 CL |
1036 | init_tracking(s, object); |
1037 | } | |
1038 | ||
1537066c | 1039 | static noinline int alloc_debug_processing(struct kmem_cache *s, struct page *page, |
ce71e27c | 1040 | void *object, unsigned long addr) |
81819f0f CL |
1041 | { |
1042 | if (!check_slab(s, page)) | |
1043 | goto bad; | |
1044 | ||
81819f0f CL |
1045 | if (!check_valid_pointer(s, page, object)) { |
1046 | object_err(s, page, object, "Freelist Pointer check fails"); | |
70d71228 | 1047 | goto bad; |
81819f0f CL |
1048 | } |
1049 | ||
f7cb1933 | 1050 | if (!check_object(s, page, object, SLUB_RED_INACTIVE)) |
81819f0f | 1051 | goto bad; |
81819f0f | 1052 | |
3ec09742 CL |
1053 | /* Success perform special debug activities for allocs */ |
1054 | if (s->flags & SLAB_STORE_USER) | |
1055 | set_track(s, object, TRACK_ALLOC, addr); | |
1056 | trace(s, page, object, 1); | |
f7cb1933 | 1057 | init_object(s, object, SLUB_RED_ACTIVE); |
81819f0f | 1058 | return 1; |
3ec09742 | 1059 | |
81819f0f CL |
1060 | bad: |
1061 | if (PageSlab(page)) { | |
1062 | /* | |
1063 | * If this is a slab page then lets do the best we can | |
1064 | * to avoid issues in the future. Marking all objects | |
672bba3a | 1065 | * as used avoids touching the remaining objects. |
81819f0f | 1066 | */ |
24922684 | 1067 | slab_fix(s, "Marking all objects used"); |
39b26464 | 1068 | page->inuse = page->objects; |
a973e9dd | 1069 | page->freelist = NULL; |
81819f0f CL |
1070 | } |
1071 | return 0; | |
1072 | } | |
1073 | ||
1537066c CL |
1074 | static noinline int free_debug_processing(struct kmem_cache *s, |
1075 | struct page *page, void *object, unsigned long addr) | |
81819f0f | 1076 | { |
5c2e4bbb CL |
1077 | unsigned long flags; |
1078 | int rc = 0; | |
1079 | ||
1080 | local_irq_save(flags); | |
881db7fb CL |
1081 | slab_lock(page); |
1082 | ||
81819f0f CL |
1083 | if (!check_slab(s, page)) |
1084 | goto fail; | |
1085 | ||
1086 | if (!check_valid_pointer(s, page, object)) { | |
70d71228 | 1087 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
81819f0f CL |
1088 | goto fail; |
1089 | } | |
1090 | ||
1091 | if (on_freelist(s, page, object)) { | |
24922684 | 1092 | object_err(s, page, object, "Object already free"); |
81819f0f CL |
1093 | goto fail; |
1094 | } | |
1095 | ||
f7cb1933 | 1096 | if (!check_object(s, page, object, SLUB_RED_ACTIVE)) |
5c2e4bbb | 1097 | goto out; |
81819f0f CL |
1098 | |
1099 | if (unlikely(s != page->slab)) { | |
3adbefee | 1100 | if (!PageSlab(page)) { |
70d71228 CL |
1101 | slab_err(s, page, "Attempt to free object(0x%p) " |
1102 | "outside of slab", object); | |
3adbefee | 1103 | } else if (!page->slab) { |
81819f0f | 1104 | printk(KERN_ERR |
70d71228 | 1105 | "SLUB <none>: no slab for object 0x%p.\n", |
81819f0f | 1106 | object); |
70d71228 | 1107 | dump_stack(); |
06428780 | 1108 | } else |
24922684 CL |
1109 | object_err(s, page, object, |
1110 | "page slab pointer corrupt."); | |
81819f0f CL |
1111 | goto fail; |
1112 | } | |
3ec09742 | 1113 | |
3ec09742 CL |
1114 | if (s->flags & SLAB_STORE_USER) |
1115 | set_track(s, object, TRACK_FREE, addr); | |
1116 | trace(s, page, object, 0); | |
f7cb1933 | 1117 | init_object(s, object, SLUB_RED_INACTIVE); |
5c2e4bbb CL |
1118 | rc = 1; |
1119 | out: | |
881db7fb | 1120 | slab_unlock(page); |
5c2e4bbb CL |
1121 | local_irq_restore(flags); |
1122 | return rc; | |
3ec09742 | 1123 | |
81819f0f | 1124 | fail: |
24922684 | 1125 | slab_fix(s, "Object at 0x%p not freed", object); |
5c2e4bbb | 1126 | goto out; |
81819f0f CL |
1127 | } |
1128 | ||
41ecc55b CL |
1129 | static int __init setup_slub_debug(char *str) |
1130 | { | |
f0630fff CL |
1131 | slub_debug = DEBUG_DEFAULT_FLAGS; |
1132 | if (*str++ != '=' || !*str) | |
1133 | /* | |
1134 | * No options specified. Switch on full debugging. | |
1135 | */ | |
1136 | goto out; | |
1137 | ||
1138 | if (*str == ',') | |
1139 | /* | |
1140 | * No options but restriction on slabs. This means full | |
1141 | * debugging for slabs matching a pattern. | |
1142 | */ | |
1143 | goto check_slabs; | |
1144 | ||
fa5ec8a1 DR |
1145 | if (tolower(*str) == 'o') { |
1146 | /* | |
1147 | * Avoid enabling debugging on caches if its minimum order | |
1148 | * would increase as a result. | |
1149 | */ | |
1150 | disable_higher_order_debug = 1; | |
1151 | goto out; | |
1152 | } | |
1153 | ||
f0630fff CL |
1154 | slub_debug = 0; |
1155 | if (*str == '-') | |
1156 | /* | |
1157 | * Switch off all debugging measures. | |
1158 | */ | |
1159 | goto out; | |
1160 | ||
1161 | /* | |
1162 | * Determine which debug features should be switched on | |
1163 | */ | |
06428780 | 1164 | for (; *str && *str != ','; str++) { |
f0630fff CL |
1165 | switch (tolower(*str)) { |
1166 | case 'f': | |
1167 | slub_debug |= SLAB_DEBUG_FREE; | |
1168 | break; | |
1169 | case 'z': | |
1170 | slub_debug |= SLAB_RED_ZONE; | |
1171 | break; | |
1172 | case 'p': | |
1173 | slub_debug |= SLAB_POISON; | |
1174 | break; | |
1175 | case 'u': | |
1176 | slub_debug |= SLAB_STORE_USER; | |
1177 | break; | |
1178 | case 't': | |
1179 | slub_debug |= SLAB_TRACE; | |
1180 | break; | |
4c13dd3b DM |
1181 | case 'a': |
1182 | slub_debug |= SLAB_FAILSLAB; | |
1183 | break; | |
f0630fff CL |
1184 | default: |
1185 | printk(KERN_ERR "slub_debug option '%c' " | |
06428780 | 1186 | "unknown. skipped\n", *str); |
f0630fff | 1187 | } |
41ecc55b CL |
1188 | } |
1189 | ||
f0630fff | 1190 | check_slabs: |
41ecc55b CL |
1191 | if (*str == ',') |
1192 | slub_debug_slabs = str + 1; | |
f0630fff | 1193 | out: |
41ecc55b CL |
1194 | return 1; |
1195 | } | |
1196 | ||
1197 | __setup("slub_debug", setup_slub_debug); | |
1198 | ||
3b0efdfa | 1199 | static unsigned long kmem_cache_flags(unsigned long object_size, |
ba0268a8 | 1200 | unsigned long flags, const char *name, |
51cc5068 | 1201 | void (*ctor)(void *)) |
41ecc55b CL |
1202 | { |
1203 | /* | |
e153362a | 1204 | * Enable debugging if selected on the kernel commandline. |
41ecc55b | 1205 | */ |
e153362a | 1206 | if (slub_debug && (!slub_debug_slabs || |
3de47213 DR |
1207 | !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs)))) |
1208 | flags |= slub_debug; | |
ba0268a8 CL |
1209 | |
1210 | return flags; | |
41ecc55b CL |
1211 | } |
1212 | #else | |
3ec09742 CL |
1213 | static inline void setup_object_debug(struct kmem_cache *s, |
1214 | struct page *page, void *object) {} | |
41ecc55b | 1215 | |
3ec09742 | 1216 | static inline int alloc_debug_processing(struct kmem_cache *s, |
ce71e27c | 1217 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1218 | |
3ec09742 | 1219 | static inline int free_debug_processing(struct kmem_cache *s, |
ce71e27c | 1220 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1221 | |
41ecc55b CL |
1222 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1223 | { return 1; } | |
1224 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 1225 | void *object, u8 val) { return 1; } |
5cc6eee8 CL |
1226 | static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1227 | struct page *page) {} | |
2cfb7455 | 1228 | static inline void remove_full(struct kmem_cache *s, struct page *page) {} |
3b0efdfa | 1229 | static inline unsigned long kmem_cache_flags(unsigned long object_size, |
ba0268a8 | 1230 | unsigned long flags, const char *name, |
51cc5068 | 1231 | void (*ctor)(void *)) |
ba0268a8 CL |
1232 | { |
1233 | return flags; | |
1234 | } | |
41ecc55b | 1235 | #define slub_debug 0 |
0f389ec6 | 1236 | |
fdaa45e9 IM |
1237 | #define disable_higher_order_debug 0 |
1238 | ||
0f389ec6 CL |
1239 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) |
1240 | { return 0; } | |
26c02cf0 AB |
1241 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1242 | { return 0; } | |
205ab99d CL |
1243 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1244 | int objects) {} | |
1245 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1246 | int objects) {} | |
7d550c56 CL |
1247 | |
1248 | static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags) | |
1249 | { return 0; } | |
1250 | ||
1251 | static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, | |
1252 | void *object) {} | |
1253 | ||
1254 | static inline void slab_free_hook(struct kmem_cache *s, void *x) {} | |
1255 | ||
ab4d5ed5 | 1256 | #endif /* CONFIG_SLUB_DEBUG */ |
205ab99d | 1257 | |
81819f0f CL |
1258 | /* |
1259 | * Slab allocation and freeing | |
1260 | */ | |
65c3376a CL |
1261 | static inline struct page *alloc_slab_page(gfp_t flags, int node, |
1262 | struct kmem_cache_order_objects oo) | |
1263 | { | |
1264 | int order = oo_order(oo); | |
1265 | ||
b1eeab67 VN |
1266 | flags |= __GFP_NOTRACK; |
1267 | ||
2154a336 | 1268 | if (node == NUMA_NO_NODE) |
65c3376a CL |
1269 | return alloc_pages(flags, order); |
1270 | else | |
6b65aaf3 | 1271 | return alloc_pages_exact_node(node, flags, order); |
65c3376a CL |
1272 | } |
1273 | ||
81819f0f CL |
1274 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
1275 | { | |
06428780 | 1276 | struct page *page; |
834f3d11 | 1277 | struct kmem_cache_order_objects oo = s->oo; |
ba52270d | 1278 | gfp_t alloc_gfp; |
81819f0f | 1279 | |
7e0528da CL |
1280 | flags &= gfp_allowed_mask; |
1281 | ||
1282 | if (flags & __GFP_WAIT) | |
1283 | local_irq_enable(); | |
1284 | ||
b7a49f0d | 1285 | flags |= s->allocflags; |
e12ba74d | 1286 | |
ba52270d PE |
1287 | /* |
1288 | * Let the initial higher-order allocation fail under memory pressure | |
1289 | * so we fall-back to the minimum order allocation. | |
1290 | */ | |
1291 | alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL; | |
1292 | ||
1293 | page = alloc_slab_page(alloc_gfp, node, oo); | |
65c3376a CL |
1294 | if (unlikely(!page)) { |
1295 | oo = s->min; | |
1296 | /* | |
1297 | * Allocation may have failed due to fragmentation. | |
1298 | * Try a lower order alloc if possible | |
1299 | */ | |
1300 | page = alloc_slab_page(flags, node, oo); | |
81819f0f | 1301 | |
7e0528da CL |
1302 | if (page) |
1303 | stat(s, ORDER_FALLBACK); | |
65c3376a | 1304 | } |
5a896d9e | 1305 | |
7e0528da CL |
1306 | if (flags & __GFP_WAIT) |
1307 | local_irq_disable(); | |
1308 | ||
1309 | if (!page) | |
1310 | return NULL; | |
1311 | ||
5a896d9e | 1312 | if (kmemcheck_enabled |
5086c389 | 1313 | && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) { |
b1eeab67 VN |
1314 | int pages = 1 << oo_order(oo); |
1315 | ||
1316 | kmemcheck_alloc_shadow(page, oo_order(oo), flags, node); | |
1317 | ||
1318 | /* | |
1319 | * Objects from caches that have a constructor don't get | |
1320 | * cleared when they're allocated, so we need to do it here. | |
1321 | */ | |
1322 | if (s->ctor) | |
1323 | kmemcheck_mark_uninitialized_pages(page, pages); | |
1324 | else | |
1325 | kmemcheck_mark_unallocated_pages(page, pages); | |
5a896d9e VN |
1326 | } |
1327 | ||
834f3d11 | 1328 | page->objects = oo_objects(oo); |
81819f0f CL |
1329 | mod_zone_page_state(page_zone(page), |
1330 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1331 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
65c3376a | 1332 | 1 << oo_order(oo)); |
81819f0f CL |
1333 | |
1334 | return page; | |
1335 | } | |
1336 | ||
1337 | static void setup_object(struct kmem_cache *s, struct page *page, | |
1338 | void *object) | |
1339 | { | |
3ec09742 | 1340 | setup_object_debug(s, page, object); |
4f104934 | 1341 | if (unlikely(s->ctor)) |
51cc5068 | 1342 | s->ctor(object); |
81819f0f CL |
1343 | } |
1344 | ||
1345 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) | |
1346 | { | |
1347 | struct page *page; | |
81819f0f | 1348 | void *start; |
81819f0f CL |
1349 | void *last; |
1350 | void *p; | |
1351 | ||
6cb06229 | 1352 | BUG_ON(flags & GFP_SLAB_BUG_MASK); |
81819f0f | 1353 | |
6cb06229 CL |
1354 | page = allocate_slab(s, |
1355 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
81819f0f CL |
1356 | if (!page) |
1357 | goto out; | |
1358 | ||
205ab99d | 1359 | inc_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f | 1360 | page->slab = s; |
c03f94cc | 1361 | __SetPageSlab(page); |
81819f0f CL |
1362 | |
1363 | start = page_address(page); | |
81819f0f CL |
1364 | |
1365 | if (unlikely(s->flags & SLAB_POISON)) | |
834f3d11 | 1366 | memset(start, POISON_INUSE, PAGE_SIZE << compound_order(page)); |
81819f0f CL |
1367 | |
1368 | last = start; | |
224a88be | 1369 | for_each_object(p, s, start, page->objects) { |
81819f0f CL |
1370 | setup_object(s, page, last); |
1371 | set_freepointer(s, last, p); | |
1372 | last = p; | |
1373 | } | |
1374 | setup_object(s, page, last); | |
a973e9dd | 1375 | set_freepointer(s, last, NULL); |
81819f0f CL |
1376 | |
1377 | page->freelist = start; | |
e6e82ea1 | 1378 | page->inuse = page->objects; |
8cb0a506 | 1379 | page->frozen = 1; |
81819f0f | 1380 | out: |
81819f0f CL |
1381 | return page; |
1382 | } | |
1383 | ||
1384 | static void __free_slab(struct kmem_cache *s, struct page *page) | |
1385 | { | |
834f3d11 CL |
1386 | int order = compound_order(page); |
1387 | int pages = 1 << order; | |
81819f0f | 1388 | |
af537b0a | 1389 | if (kmem_cache_debug(s)) { |
81819f0f CL |
1390 | void *p; |
1391 | ||
1392 | slab_pad_check(s, page); | |
224a88be CL |
1393 | for_each_object(p, s, page_address(page), |
1394 | page->objects) | |
f7cb1933 | 1395 | check_object(s, page, p, SLUB_RED_INACTIVE); |
81819f0f CL |
1396 | } |
1397 | ||
b1eeab67 | 1398 | kmemcheck_free_shadow(page, compound_order(page)); |
5a896d9e | 1399 | |
81819f0f CL |
1400 | mod_zone_page_state(page_zone(page), |
1401 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1402 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
06428780 | 1403 | -pages); |
81819f0f | 1404 | |
49bd5221 CL |
1405 | __ClearPageSlab(page); |
1406 | reset_page_mapcount(page); | |
1eb5ac64 NP |
1407 | if (current->reclaim_state) |
1408 | current->reclaim_state->reclaimed_slab += pages; | |
834f3d11 | 1409 | __free_pages(page, order); |
81819f0f CL |
1410 | } |
1411 | ||
da9a638c LJ |
1412 | #define need_reserve_slab_rcu \ |
1413 | (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head)) | |
1414 | ||
81819f0f CL |
1415 | static void rcu_free_slab(struct rcu_head *h) |
1416 | { | |
1417 | struct page *page; | |
1418 | ||
da9a638c LJ |
1419 | if (need_reserve_slab_rcu) |
1420 | page = virt_to_head_page(h); | |
1421 | else | |
1422 | page = container_of((struct list_head *)h, struct page, lru); | |
1423 | ||
81819f0f CL |
1424 | __free_slab(page->slab, page); |
1425 | } | |
1426 | ||
1427 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1428 | { | |
1429 | if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) { | |
da9a638c LJ |
1430 | struct rcu_head *head; |
1431 | ||
1432 | if (need_reserve_slab_rcu) { | |
1433 | int order = compound_order(page); | |
1434 | int offset = (PAGE_SIZE << order) - s->reserved; | |
1435 | ||
1436 | VM_BUG_ON(s->reserved != sizeof(*head)); | |
1437 | head = page_address(page) + offset; | |
1438 | } else { | |
1439 | /* | |
1440 | * RCU free overloads the RCU head over the LRU | |
1441 | */ | |
1442 | head = (void *)&page->lru; | |
1443 | } | |
81819f0f CL |
1444 | |
1445 | call_rcu(head, rcu_free_slab); | |
1446 | } else | |
1447 | __free_slab(s, page); | |
1448 | } | |
1449 | ||
1450 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1451 | { | |
205ab99d | 1452 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1453 | free_slab(s, page); |
1454 | } | |
1455 | ||
1456 | /* | |
5cc6eee8 CL |
1457 | * Management of partially allocated slabs. |
1458 | * | |
1459 | * list_lock must be held. | |
81819f0f | 1460 | */ |
5cc6eee8 | 1461 | static inline void add_partial(struct kmem_cache_node *n, |
7c2e132c | 1462 | struct page *page, int tail) |
81819f0f | 1463 | { |
e95eed57 | 1464 | n->nr_partial++; |
136333d1 | 1465 | if (tail == DEACTIVATE_TO_TAIL) |
7c2e132c CL |
1466 | list_add_tail(&page->lru, &n->partial); |
1467 | else | |
1468 | list_add(&page->lru, &n->partial); | |
81819f0f CL |
1469 | } |
1470 | ||
5cc6eee8 CL |
1471 | /* |
1472 | * list_lock must be held. | |
1473 | */ | |
1474 | static inline void remove_partial(struct kmem_cache_node *n, | |
62e346a8 CL |
1475 | struct page *page) |
1476 | { | |
1477 | list_del(&page->lru); | |
1478 | n->nr_partial--; | |
1479 | } | |
1480 | ||
81819f0f | 1481 | /* |
7ced3719 CL |
1482 | * Remove slab from the partial list, freeze it and |
1483 | * return the pointer to the freelist. | |
81819f0f | 1484 | * |
497b66f2 CL |
1485 | * Returns a list of objects or NULL if it fails. |
1486 | * | |
7ced3719 | 1487 | * Must hold list_lock since we modify the partial list. |
81819f0f | 1488 | */ |
497b66f2 | 1489 | static inline void *acquire_slab(struct kmem_cache *s, |
acd19fd1 | 1490 | struct kmem_cache_node *n, struct page *page, |
49e22585 | 1491 | int mode) |
81819f0f | 1492 | { |
2cfb7455 CL |
1493 | void *freelist; |
1494 | unsigned long counters; | |
1495 | struct page new; | |
1496 | ||
2cfb7455 CL |
1497 | /* |
1498 | * Zap the freelist and set the frozen bit. | |
1499 | * The old freelist is the list of objects for the | |
1500 | * per cpu allocation list. | |
1501 | */ | |
7ced3719 CL |
1502 | freelist = page->freelist; |
1503 | counters = page->counters; | |
1504 | new.counters = counters; | |
23910c50 | 1505 | if (mode) { |
7ced3719 | 1506 | new.inuse = page->objects; |
23910c50 PE |
1507 | new.freelist = NULL; |
1508 | } else { | |
1509 | new.freelist = freelist; | |
1510 | } | |
2cfb7455 | 1511 | |
7ced3719 CL |
1512 | VM_BUG_ON(new.frozen); |
1513 | new.frozen = 1; | |
2cfb7455 | 1514 | |
7ced3719 | 1515 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 | 1516 | freelist, counters, |
02d7633f | 1517 | new.freelist, new.counters, |
7ced3719 | 1518 | "acquire_slab")) |
7ced3719 | 1519 | return NULL; |
2cfb7455 CL |
1520 | |
1521 | remove_partial(n, page); | |
7ced3719 | 1522 | WARN_ON(!freelist); |
49e22585 | 1523 | return freelist; |
81819f0f CL |
1524 | } |
1525 | ||
49e22585 CL |
1526 | static int put_cpu_partial(struct kmem_cache *s, struct page *page, int drain); |
1527 | ||
81819f0f | 1528 | /* |
672bba3a | 1529 | * Try to allocate a partial slab from a specific node. |
81819f0f | 1530 | */ |
497b66f2 | 1531 | static void *get_partial_node(struct kmem_cache *s, |
acd19fd1 | 1532 | struct kmem_cache_node *n, struct kmem_cache_cpu *c) |
81819f0f | 1533 | { |
49e22585 CL |
1534 | struct page *page, *page2; |
1535 | void *object = NULL; | |
81819f0f CL |
1536 | |
1537 | /* | |
1538 | * Racy check. If we mistakenly see no partial slabs then we | |
1539 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1540 | * partial slab and there is none available then get_partials() |
1541 | * will return NULL. | |
81819f0f CL |
1542 | */ |
1543 | if (!n || !n->nr_partial) | |
1544 | return NULL; | |
1545 | ||
1546 | spin_lock(&n->list_lock); | |
49e22585 | 1547 | list_for_each_entry_safe(page, page2, &n->partial, lru) { |
12d79634 | 1548 | void *t = acquire_slab(s, n, page, object == NULL); |
49e22585 CL |
1549 | int available; |
1550 | ||
1551 | if (!t) | |
1552 | break; | |
1553 | ||
12d79634 | 1554 | if (!object) { |
49e22585 | 1555 | c->page = page; |
49e22585 | 1556 | stat(s, ALLOC_FROM_PARTIAL); |
49e22585 CL |
1557 | object = t; |
1558 | available = page->objects - page->inuse; | |
1559 | } else { | |
49e22585 | 1560 | available = put_cpu_partial(s, page, 0); |
8028dcea | 1561 | stat(s, CPU_PARTIAL_NODE); |
49e22585 CL |
1562 | } |
1563 | if (kmem_cache_debug(s) || available > s->cpu_partial / 2) | |
1564 | break; | |
1565 | ||
497b66f2 | 1566 | } |
81819f0f | 1567 | spin_unlock(&n->list_lock); |
497b66f2 | 1568 | return object; |
81819f0f CL |
1569 | } |
1570 | ||
1571 | /* | |
672bba3a | 1572 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f | 1573 | */ |
de3ec035 | 1574 | static void *get_any_partial(struct kmem_cache *s, gfp_t flags, |
acd19fd1 | 1575 | struct kmem_cache_cpu *c) |
81819f0f CL |
1576 | { |
1577 | #ifdef CONFIG_NUMA | |
1578 | struct zonelist *zonelist; | |
dd1a239f | 1579 | struct zoneref *z; |
54a6eb5c MG |
1580 | struct zone *zone; |
1581 | enum zone_type high_zoneidx = gfp_zone(flags); | |
497b66f2 | 1582 | void *object; |
cc9a6c87 | 1583 | unsigned int cpuset_mems_cookie; |
81819f0f CL |
1584 | |
1585 | /* | |
672bba3a CL |
1586 | * The defrag ratio allows a configuration of the tradeoffs between |
1587 | * inter node defragmentation and node local allocations. A lower | |
1588 | * defrag_ratio increases the tendency to do local allocations | |
1589 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1590 | * |
672bba3a CL |
1591 | * If the defrag_ratio is set to 0 then kmalloc() always |
1592 | * returns node local objects. If the ratio is higher then kmalloc() | |
1593 | * may return off node objects because partial slabs are obtained | |
1594 | * from other nodes and filled up. | |
81819f0f | 1595 | * |
6446faa2 | 1596 | * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes |
672bba3a CL |
1597 | * defrag_ratio = 1000) then every (well almost) allocation will |
1598 | * first attempt to defrag slab caches on other nodes. This means | |
1599 | * scanning over all nodes to look for partial slabs which may be | |
1600 | * expensive if we do it every time we are trying to find a slab | |
1601 | * with available objects. | |
81819f0f | 1602 | */ |
9824601e CL |
1603 | if (!s->remote_node_defrag_ratio || |
1604 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
1605 | return NULL; |
1606 | ||
cc9a6c87 MG |
1607 | do { |
1608 | cpuset_mems_cookie = get_mems_allowed(); | |
e7b691b0 | 1609 | zonelist = node_zonelist(slab_node(), flags); |
cc9a6c87 MG |
1610 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
1611 | struct kmem_cache_node *n; | |
1612 | ||
1613 | n = get_node(s, zone_to_nid(zone)); | |
1614 | ||
1615 | if (n && cpuset_zone_allowed_hardwall(zone, flags) && | |
1616 | n->nr_partial > s->min_partial) { | |
1617 | object = get_partial_node(s, n, c); | |
1618 | if (object) { | |
1619 | /* | |
1620 | * Return the object even if | |
1621 | * put_mems_allowed indicated that | |
1622 | * the cpuset mems_allowed was | |
1623 | * updated in parallel. It's a | |
1624 | * harmless race between the alloc | |
1625 | * and the cpuset update. | |
1626 | */ | |
1627 | put_mems_allowed(cpuset_mems_cookie); | |
1628 | return object; | |
1629 | } | |
c0ff7453 | 1630 | } |
81819f0f | 1631 | } |
cc9a6c87 | 1632 | } while (!put_mems_allowed(cpuset_mems_cookie)); |
81819f0f CL |
1633 | #endif |
1634 | return NULL; | |
1635 | } | |
1636 | ||
1637 | /* | |
1638 | * Get a partial page, lock it and return it. | |
1639 | */ | |
497b66f2 | 1640 | static void *get_partial(struct kmem_cache *s, gfp_t flags, int node, |
acd19fd1 | 1641 | struct kmem_cache_cpu *c) |
81819f0f | 1642 | { |
497b66f2 | 1643 | void *object; |
2154a336 | 1644 | int searchnode = (node == NUMA_NO_NODE) ? numa_node_id() : node; |
81819f0f | 1645 | |
497b66f2 CL |
1646 | object = get_partial_node(s, get_node(s, searchnode), c); |
1647 | if (object || node != NUMA_NO_NODE) | |
1648 | return object; | |
81819f0f | 1649 | |
acd19fd1 | 1650 | return get_any_partial(s, flags, c); |
81819f0f CL |
1651 | } |
1652 | ||
8a5ec0ba CL |
1653 | #ifdef CONFIG_PREEMPT |
1654 | /* | |
1655 | * Calculate the next globally unique transaction for disambiguiation | |
1656 | * during cmpxchg. The transactions start with the cpu number and are then | |
1657 | * incremented by CONFIG_NR_CPUS. | |
1658 | */ | |
1659 | #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS) | |
1660 | #else | |
1661 | /* | |
1662 | * No preemption supported therefore also no need to check for | |
1663 | * different cpus. | |
1664 | */ | |
1665 | #define TID_STEP 1 | |
1666 | #endif | |
1667 | ||
1668 | static inline unsigned long next_tid(unsigned long tid) | |
1669 | { | |
1670 | return tid + TID_STEP; | |
1671 | } | |
1672 | ||
1673 | static inline unsigned int tid_to_cpu(unsigned long tid) | |
1674 | { | |
1675 | return tid % TID_STEP; | |
1676 | } | |
1677 | ||
1678 | static inline unsigned long tid_to_event(unsigned long tid) | |
1679 | { | |
1680 | return tid / TID_STEP; | |
1681 | } | |
1682 | ||
1683 | static inline unsigned int init_tid(int cpu) | |
1684 | { | |
1685 | return cpu; | |
1686 | } | |
1687 | ||
1688 | static inline void note_cmpxchg_failure(const char *n, | |
1689 | const struct kmem_cache *s, unsigned long tid) | |
1690 | { | |
1691 | #ifdef SLUB_DEBUG_CMPXCHG | |
1692 | unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid); | |
1693 | ||
1694 | printk(KERN_INFO "%s %s: cmpxchg redo ", n, s->name); | |
1695 | ||
1696 | #ifdef CONFIG_PREEMPT | |
1697 | if (tid_to_cpu(tid) != tid_to_cpu(actual_tid)) | |
1698 | printk("due to cpu change %d -> %d\n", | |
1699 | tid_to_cpu(tid), tid_to_cpu(actual_tid)); | |
1700 | else | |
1701 | #endif | |
1702 | if (tid_to_event(tid) != tid_to_event(actual_tid)) | |
1703 | printk("due to cpu running other code. Event %ld->%ld\n", | |
1704 | tid_to_event(tid), tid_to_event(actual_tid)); | |
1705 | else | |
1706 | printk("for unknown reason: actual=%lx was=%lx target=%lx\n", | |
1707 | actual_tid, tid, next_tid(tid)); | |
1708 | #endif | |
4fdccdfb | 1709 | stat(s, CMPXCHG_DOUBLE_CPU_FAIL); |
8a5ec0ba CL |
1710 | } |
1711 | ||
8a5ec0ba CL |
1712 | void init_kmem_cache_cpus(struct kmem_cache *s) |
1713 | { | |
8a5ec0ba CL |
1714 | int cpu; |
1715 | ||
1716 | for_each_possible_cpu(cpu) | |
1717 | per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu); | |
8a5ec0ba | 1718 | } |
2cfb7455 | 1719 | |
81819f0f CL |
1720 | /* |
1721 | * Remove the cpu slab | |
1722 | */ | |
c17dda40 | 1723 | static void deactivate_slab(struct kmem_cache *s, struct page *page, void *freelist) |
81819f0f | 1724 | { |
2cfb7455 | 1725 | enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE }; |
2cfb7455 CL |
1726 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
1727 | int lock = 0; | |
1728 | enum slab_modes l = M_NONE, m = M_NONE; | |
2cfb7455 | 1729 | void *nextfree; |
136333d1 | 1730 | int tail = DEACTIVATE_TO_HEAD; |
2cfb7455 CL |
1731 | struct page new; |
1732 | struct page old; | |
1733 | ||
1734 | if (page->freelist) { | |
84e554e6 | 1735 | stat(s, DEACTIVATE_REMOTE_FREES); |
136333d1 | 1736 | tail = DEACTIVATE_TO_TAIL; |
2cfb7455 CL |
1737 | } |
1738 | ||
894b8788 | 1739 | /* |
2cfb7455 CL |
1740 | * Stage one: Free all available per cpu objects back |
1741 | * to the page freelist while it is still frozen. Leave the | |
1742 | * last one. | |
1743 | * | |
1744 | * There is no need to take the list->lock because the page | |
1745 | * is still frozen. | |
1746 | */ | |
1747 | while (freelist && (nextfree = get_freepointer(s, freelist))) { | |
1748 | void *prior; | |
1749 | unsigned long counters; | |
1750 | ||
1751 | do { | |
1752 | prior = page->freelist; | |
1753 | counters = page->counters; | |
1754 | set_freepointer(s, freelist, prior); | |
1755 | new.counters = counters; | |
1756 | new.inuse--; | |
1757 | VM_BUG_ON(!new.frozen); | |
1758 | ||
1d07171c | 1759 | } while (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
1760 | prior, counters, |
1761 | freelist, new.counters, | |
1762 | "drain percpu freelist")); | |
1763 | ||
1764 | freelist = nextfree; | |
1765 | } | |
1766 | ||
894b8788 | 1767 | /* |
2cfb7455 CL |
1768 | * Stage two: Ensure that the page is unfrozen while the |
1769 | * list presence reflects the actual number of objects | |
1770 | * during unfreeze. | |
1771 | * | |
1772 | * We setup the list membership and then perform a cmpxchg | |
1773 | * with the count. If there is a mismatch then the page | |
1774 | * is not unfrozen but the page is on the wrong list. | |
1775 | * | |
1776 | * Then we restart the process which may have to remove | |
1777 | * the page from the list that we just put it on again | |
1778 | * because the number of objects in the slab may have | |
1779 | * changed. | |
894b8788 | 1780 | */ |
2cfb7455 | 1781 | redo: |
894b8788 | 1782 | |
2cfb7455 CL |
1783 | old.freelist = page->freelist; |
1784 | old.counters = page->counters; | |
1785 | VM_BUG_ON(!old.frozen); | |
7c2e132c | 1786 | |
2cfb7455 CL |
1787 | /* Determine target state of the slab */ |
1788 | new.counters = old.counters; | |
1789 | if (freelist) { | |
1790 | new.inuse--; | |
1791 | set_freepointer(s, freelist, old.freelist); | |
1792 | new.freelist = freelist; | |
1793 | } else | |
1794 | new.freelist = old.freelist; | |
1795 | ||
1796 | new.frozen = 0; | |
1797 | ||
81107188 | 1798 | if (!new.inuse && n->nr_partial > s->min_partial) |
2cfb7455 CL |
1799 | m = M_FREE; |
1800 | else if (new.freelist) { | |
1801 | m = M_PARTIAL; | |
1802 | if (!lock) { | |
1803 | lock = 1; | |
1804 | /* | |
1805 | * Taking the spinlock removes the possiblity | |
1806 | * that acquire_slab() will see a slab page that | |
1807 | * is frozen | |
1808 | */ | |
1809 | spin_lock(&n->list_lock); | |
1810 | } | |
1811 | } else { | |
1812 | m = M_FULL; | |
1813 | if (kmem_cache_debug(s) && !lock) { | |
1814 | lock = 1; | |
1815 | /* | |
1816 | * This also ensures that the scanning of full | |
1817 | * slabs from diagnostic functions will not see | |
1818 | * any frozen slabs. | |
1819 | */ | |
1820 | spin_lock(&n->list_lock); | |
1821 | } | |
1822 | } | |
1823 | ||
1824 | if (l != m) { | |
1825 | ||
1826 | if (l == M_PARTIAL) | |
1827 | ||
1828 | remove_partial(n, page); | |
1829 | ||
1830 | else if (l == M_FULL) | |
894b8788 | 1831 | |
2cfb7455 CL |
1832 | remove_full(s, page); |
1833 | ||
1834 | if (m == M_PARTIAL) { | |
1835 | ||
1836 | add_partial(n, page, tail); | |
136333d1 | 1837 | stat(s, tail); |
2cfb7455 CL |
1838 | |
1839 | } else if (m == M_FULL) { | |
894b8788 | 1840 | |
2cfb7455 CL |
1841 | stat(s, DEACTIVATE_FULL); |
1842 | add_full(s, n, page); | |
1843 | ||
1844 | } | |
1845 | } | |
1846 | ||
1847 | l = m; | |
1d07171c | 1848 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
1849 | old.freelist, old.counters, |
1850 | new.freelist, new.counters, | |
1851 | "unfreezing slab")) | |
1852 | goto redo; | |
1853 | ||
2cfb7455 CL |
1854 | if (lock) |
1855 | spin_unlock(&n->list_lock); | |
1856 | ||
1857 | if (m == M_FREE) { | |
1858 | stat(s, DEACTIVATE_EMPTY); | |
1859 | discard_slab(s, page); | |
1860 | stat(s, FREE_SLAB); | |
894b8788 | 1861 | } |
81819f0f CL |
1862 | } |
1863 | ||
d24ac77f JK |
1864 | /* |
1865 | * Unfreeze all the cpu partial slabs. | |
1866 | * | |
1867 | * This function must be called with interrupt disabled. | |
1868 | */ | |
49e22585 CL |
1869 | static void unfreeze_partials(struct kmem_cache *s) |
1870 | { | |
43d77867 | 1871 | struct kmem_cache_node *n = NULL, *n2 = NULL; |
49e22585 | 1872 | struct kmem_cache_cpu *c = this_cpu_ptr(s->cpu_slab); |
9ada1934 | 1873 | struct page *page, *discard_page = NULL; |
49e22585 CL |
1874 | |
1875 | while ((page = c->partial)) { | |
49e22585 CL |
1876 | struct page new; |
1877 | struct page old; | |
1878 | ||
1879 | c->partial = page->next; | |
43d77867 JK |
1880 | |
1881 | n2 = get_node(s, page_to_nid(page)); | |
1882 | if (n != n2) { | |
1883 | if (n) | |
1884 | spin_unlock(&n->list_lock); | |
1885 | ||
1886 | n = n2; | |
1887 | spin_lock(&n->list_lock); | |
1888 | } | |
49e22585 CL |
1889 | |
1890 | do { | |
1891 | ||
1892 | old.freelist = page->freelist; | |
1893 | old.counters = page->counters; | |
1894 | VM_BUG_ON(!old.frozen); | |
1895 | ||
1896 | new.counters = old.counters; | |
1897 | new.freelist = old.freelist; | |
1898 | ||
1899 | new.frozen = 0; | |
1900 | ||
d24ac77f | 1901 | } while (!__cmpxchg_double_slab(s, page, |
49e22585 CL |
1902 | old.freelist, old.counters, |
1903 | new.freelist, new.counters, | |
1904 | "unfreezing slab")); | |
1905 | ||
43d77867 | 1906 | if (unlikely(!new.inuse && n->nr_partial > s->min_partial)) { |
9ada1934 SL |
1907 | page->next = discard_page; |
1908 | discard_page = page; | |
43d77867 JK |
1909 | } else { |
1910 | add_partial(n, page, DEACTIVATE_TO_TAIL); | |
1911 | stat(s, FREE_ADD_PARTIAL); | |
49e22585 CL |
1912 | } |
1913 | } | |
1914 | ||
1915 | if (n) | |
1916 | spin_unlock(&n->list_lock); | |
9ada1934 SL |
1917 | |
1918 | while (discard_page) { | |
1919 | page = discard_page; | |
1920 | discard_page = discard_page->next; | |
1921 | ||
1922 | stat(s, DEACTIVATE_EMPTY); | |
1923 | discard_slab(s, page); | |
1924 | stat(s, FREE_SLAB); | |
1925 | } | |
49e22585 CL |
1926 | } |
1927 | ||
1928 | /* | |
1929 | * Put a page that was just frozen (in __slab_free) into a partial page | |
1930 | * slot if available. This is done without interrupts disabled and without | |
1931 | * preemption disabled. The cmpxchg is racy and may put the partial page | |
1932 | * onto a random cpus partial slot. | |
1933 | * | |
1934 | * If we did not find a slot then simply move all the partials to the | |
1935 | * per node partial list. | |
1936 | */ | |
1937 | int put_cpu_partial(struct kmem_cache *s, struct page *page, int drain) | |
1938 | { | |
1939 | struct page *oldpage; | |
1940 | int pages; | |
1941 | int pobjects; | |
1942 | ||
1943 | do { | |
1944 | pages = 0; | |
1945 | pobjects = 0; | |
1946 | oldpage = this_cpu_read(s->cpu_slab->partial); | |
1947 | ||
1948 | if (oldpage) { | |
1949 | pobjects = oldpage->pobjects; | |
1950 | pages = oldpage->pages; | |
1951 | if (drain && pobjects > s->cpu_partial) { | |
1952 | unsigned long flags; | |
1953 | /* | |
1954 | * partial array is full. Move the existing | |
1955 | * set to the per node partial list. | |
1956 | */ | |
1957 | local_irq_save(flags); | |
1958 | unfreeze_partials(s); | |
1959 | local_irq_restore(flags); | |
1960 | pobjects = 0; | |
1961 | pages = 0; | |
8028dcea | 1962 | stat(s, CPU_PARTIAL_DRAIN); |
49e22585 CL |
1963 | } |
1964 | } | |
1965 | ||
1966 | pages++; | |
1967 | pobjects += page->objects - page->inuse; | |
1968 | ||
1969 | page->pages = pages; | |
1970 | page->pobjects = pobjects; | |
1971 | page->next = oldpage; | |
1972 | ||
933393f5 | 1973 | } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page) != oldpage); |
49e22585 CL |
1974 | return pobjects; |
1975 | } | |
1976 | ||
dfb4f096 | 1977 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 1978 | { |
84e554e6 | 1979 | stat(s, CPUSLAB_FLUSH); |
c17dda40 CL |
1980 | deactivate_slab(s, c->page, c->freelist); |
1981 | ||
1982 | c->tid = next_tid(c->tid); | |
1983 | c->page = NULL; | |
1984 | c->freelist = NULL; | |
81819f0f CL |
1985 | } |
1986 | ||
1987 | /* | |
1988 | * Flush cpu slab. | |
6446faa2 | 1989 | * |
81819f0f CL |
1990 | * Called from IPI handler with interrupts disabled. |
1991 | */ | |
0c710013 | 1992 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 1993 | { |
9dfc6e68 | 1994 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
81819f0f | 1995 | |
49e22585 CL |
1996 | if (likely(c)) { |
1997 | if (c->page) | |
1998 | flush_slab(s, c); | |
1999 | ||
2000 | unfreeze_partials(s); | |
2001 | } | |
81819f0f CL |
2002 | } |
2003 | ||
2004 | static void flush_cpu_slab(void *d) | |
2005 | { | |
2006 | struct kmem_cache *s = d; | |
81819f0f | 2007 | |
dfb4f096 | 2008 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
2009 | } |
2010 | ||
a8364d55 GBY |
2011 | static bool has_cpu_slab(int cpu, void *info) |
2012 | { | |
2013 | struct kmem_cache *s = info; | |
2014 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); | |
2015 | ||
02e1a9cd | 2016 | return c->page || c->partial; |
a8364d55 GBY |
2017 | } |
2018 | ||
81819f0f CL |
2019 | static void flush_all(struct kmem_cache *s) |
2020 | { | |
a8364d55 | 2021 | on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1, GFP_ATOMIC); |
81819f0f CL |
2022 | } |
2023 | ||
dfb4f096 CL |
2024 | /* |
2025 | * Check if the objects in a per cpu structure fit numa | |
2026 | * locality expectations. | |
2027 | */ | |
57d437d2 | 2028 | static inline int node_match(struct page *page, int node) |
dfb4f096 CL |
2029 | { |
2030 | #ifdef CONFIG_NUMA | |
57d437d2 | 2031 | if (node != NUMA_NO_NODE && page_to_nid(page) != node) |
dfb4f096 CL |
2032 | return 0; |
2033 | #endif | |
2034 | return 1; | |
2035 | } | |
2036 | ||
781b2ba6 PE |
2037 | static int count_free(struct page *page) |
2038 | { | |
2039 | return page->objects - page->inuse; | |
2040 | } | |
2041 | ||
2042 | static unsigned long count_partial(struct kmem_cache_node *n, | |
2043 | int (*get_count)(struct page *)) | |
2044 | { | |
2045 | unsigned long flags; | |
2046 | unsigned long x = 0; | |
2047 | struct page *page; | |
2048 | ||
2049 | spin_lock_irqsave(&n->list_lock, flags); | |
2050 | list_for_each_entry(page, &n->partial, lru) | |
2051 | x += get_count(page); | |
2052 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2053 | return x; | |
2054 | } | |
2055 | ||
26c02cf0 AB |
2056 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
2057 | { | |
2058 | #ifdef CONFIG_SLUB_DEBUG | |
2059 | return atomic_long_read(&n->total_objects); | |
2060 | #else | |
2061 | return 0; | |
2062 | #endif | |
2063 | } | |
2064 | ||
781b2ba6 PE |
2065 | static noinline void |
2066 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
2067 | { | |
2068 | int node; | |
2069 | ||
2070 | printk(KERN_WARNING | |
2071 | "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n", | |
2072 | nid, gfpflags); | |
2073 | printk(KERN_WARNING " cache: %s, object size: %d, buffer size: %d, " | |
3b0efdfa | 2074 | "default order: %d, min order: %d\n", s->name, s->object_size, |
781b2ba6 PE |
2075 | s->size, oo_order(s->oo), oo_order(s->min)); |
2076 | ||
3b0efdfa | 2077 | if (oo_order(s->min) > get_order(s->object_size)) |
fa5ec8a1 DR |
2078 | printk(KERN_WARNING " %s debugging increased min order, use " |
2079 | "slub_debug=O to disable.\n", s->name); | |
2080 | ||
781b2ba6 PE |
2081 | for_each_online_node(node) { |
2082 | struct kmem_cache_node *n = get_node(s, node); | |
2083 | unsigned long nr_slabs; | |
2084 | unsigned long nr_objs; | |
2085 | unsigned long nr_free; | |
2086 | ||
2087 | if (!n) | |
2088 | continue; | |
2089 | ||
26c02cf0 AB |
2090 | nr_free = count_partial(n, count_free); |
2091 | nr_slabs = node_nr_slabs(n); | |
2092 | nr_objs = node_nr_objs(n); | |
781b2ba6 PE |
2093 | |
2094 | printk(KERN_WARNING | |
2095 | " node %d: slabs: %ld, objs: %ld, free: %ld\n", | |
2096 | node, nr_slabs, nr_objs, nr_free); | |
2097 | } | |
2098 | } | |
2099 | ||
497b66f2 CL |
2100 | static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags, |
2101 | int node, struct kmem_cache_cpu **pc) | |
2102 | { | |
6faa6833 | 2103 | void *freelist; |
188fd063 CL |
2104 | struct kmem_cache_cpu *c = *pc; |
2105 | struct page *page; | |
497b66f2 | 2106 | |
188fd063 | 2107 | freelist = get_partial(s, flags, node, c); |
497b66f2 | 2108 | |
188fd063 CL |
2109 | if (freelist) |
2110 | return freelist; | |
2111 | ||
2112 | page = new_slab(s, flags, node); | |
497b66f2 CL |
2113 | if (page) { |
2114 | c = __this_cpu_ptr(s->cpu_slab); | |
2115 | if (c->page) | |
2116 | flush_slab(s, c); | |
2117 | ||
2118 | /* | |
2119 | * No other reference to the page yet so we can | |
2120 | * muck around with it freely without cmpxchg | |
2121 | */ | |
6faa6833 | 2122 | freelist = page->freelist; |
497b66f2 CL |
2123 | page->freelist = NULL; |
2124 | ||
2125 | stat(s, ALLOC_SLAB); | |
497b66f2 CL |
2126 | c->page = page; |
2127 | *pc = c; | |
2128 | } else | |
6faa6833 | 2129 | freelist = NULL; |
497b66f2 | 2130 | |
6faa6833 | 2131 | return freelist; |
497b66f2 CL |
2132 | } |
2133 | ||
213eeb9f CL |
2134 | /* |
2135 | * Check the page->freelist of a page and either transfer the freelist to the per cpu freelist | |
2136 | * or deactivate the page. | |
2137 | * | |
2138 | * The page is still frozen if the return value is not NULL. | |
2139 | * | |
2140 | * If this function returns NULL then the page has been unfrozen. | |
d24ac77f JK |
2141 | * |
2142 | * This function must be called with interrupt disabled. | |
213eeb9f CL |
2143 | */ |
2144 | static inline void *get_freelist(struct kmem_cache *s, struct page *page) | |
2145 | { | |
2146 | struct page new; | |
2147 | unsigned long counters; | |
2148 | void *freelist; | |
2149 | ||
2150 | do { | |
2151 | freelist = page->freelist; | |
2152 | counters = page->counters; | |
6faa6833 | 2153 | |
213eeb9f CL |
2154 | new.counters = counters; |
2155 | VM_BUG_ON(!new.frozen); | |
2156 | ||
2157 | new.inuse = page->objects; | |
2158 | new.frozen = freelist != NULL; | |
2159 | ||
d24ac77f | 2160 | } while (!__cmpxchg_double_slab(s, page, |
213eeb9f CL |
2161 | freelist, counters, |
2162 | NULL, new.counters, | |
2163 | "get_freelist")); | |
2164 | ||
2165 | return freelist; | |
2166 | } | |
2167 | ||
81819f0f | 2168 | /* |
894b8788 CL |
2169 | * Slow path. The lockless freelist is empty or we need to perform |
2170 | * debugging duties. | |
2171 | * | |
894b8788 CL |
2172 | * Processing is still very fast if new objects have been freed to the |
2173 | * regular freelist. In that case we simply take over the regular freelist | |
2174 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 2175 | * |
894b8788 CL |
2176 | * If that is not working then we fall back to the partial lists. We take the |
2177 | * first element of the freelist as the object to allocate now and move the | |
2178 | * rest of the freelist to the lockless freelist. | |
81819f0f | 2179 | * |
894b8788 | 2180 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
2181 | * we need to allocate a new slab. This is the slowest path since it involves |
2182 | * a call to the page allocator and the setup of a new slab. | |
81819f0f | 2183 | */ |
ce71e27c EGM |
2184 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
2185 | unsigned long addr, struct kmem_cache_cpu *c) | |
81819f0f | 2186 | { |
6faa6833 | 2187 | void *freelist; |
f6e7def7 | 2188 | struct page *page; |
8a5ec0ba CL |
2189 | unsigned long flags; |
2190 | ||
2191 | local_irq_save(flags); | |
2192 | #ifdef CONFIG_PREEMPT | |
2193 | /* | |
2194 | * We may have been preempted and rescheduled on a different | |
2195 | * cpu before disabling interrupts. Need to reload cpu area | |
2196 | * pointer. | |
2197 | */ | |
2198 | c = this_cpu_ptr(s->cpu_slab); | |
8a5ec0ba | 2199 | #endif |
81819f0f | 2200 | |
f6e7def7 CL |
2201 | page = c->page; |
2202 | if (!page) | |
81819f0f | 2203 | goto new_slab; |
49e22585 | 2204 | redo: |
6faa6833 | 2205 | |
57d437d2 | 2206 | if (unlikely(!node_match(page, node))) { |
e36a2652 | 2207 | stat(s, ALLOC_NODE_MISMATCH); |
f6e7def7 | 2208 | deactivate_slab(s, page, c->freelist); |
c17dda40 CL |
2209 | c->page = NULL; |
2210 | c->freelist = NULL; | |
fc59c053 CL |
2211 | goto new_slab; |
2212 | } | |
6446faa2 | 2213 | |
73736e03 | 2214 | /* must check again c->freelist in case of cpu migration or IRQ */ |
6faa6833 CL |
2215 | freelist = c->freelist; |
2216 | if (freelist) | |
73736e03 | 2217 | goto load_freelist; |
03e404af | 2218 | |
2cfb7455 | 2219 | stat(s, ALLOC_SLOWPATH); |
03e404af | 2220 | |
f6e7def7 | 2221 | freelist = get_freelist(s, page); |
6446faa2 | 2222 | |
6faa6833 | 2223 | if (!freelist) { |
03e404af CL |
2224 | c->page = NULL; |
2225 | stat(s, DEACTIVATE_BYPASS); | |
fc59c053 | 2226 | goto new_slab; |
03e404af | 2227 | } |
6446faa2 | 2228 | |
84e554e6 | 2229 | stat(s, ALLOC_REFILL); |
6446faa2 | 2230 | |
894b8788 | 2231 | load_freelist: |
507effea CL |
2232 | /* |
2233 | * freelist is pointing to the list of objects to be used. | |
2234 | * page is pointing to the page from which the objects are obtained. | |
2235 | * That page must be frozen for per cpu allocations to work. | |
2236 | */ | |
2237 | VM_BUG_ON(!c->page->frozen); | |
6faa6833 | 2238 | c->freelist = get_freepointer(s, freelist); |
8a5ec0ba CL |
2239 | c->tid = next_tid(c->tid); |
2240 | local_irq_restore(flags); | |
6faa6833 | 2241 | return freelist; |
81819f0f | 2242 | |
81819f0f | 2243 | new_slab: |
2cfb7455 | 2244 | |
49e22585 | 2245 | if (c->partial) { |
f6e7def7 CL |
2246 | page = c->page = c->partial; |
2247 | c->partial = page->next; | |
49e22585 CL |
2248 | stat(s, CPU_PARTIAL_ALLOC); |
2249 | c->freelist = NULL; | |
2250 | goto redo; | |
81819f0f CL |
2251 | } |
2252 | ||
188fd063 | 2253 | freelist = new_slab_objects(s, gfpflags, node, &c); |
01ad8a7b | 2254 | |
f4697436 CL |
2255 | if (unlikely(!freelist)) { |
2256 | if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit()) | |
2257 | slab_out_of_memory(s, gfpflags, node); | |
2cfb7455 | 2258 | |
f4697436 CL |
2259 | local_irq_restore(flags); |
2260 | return NULL; | |
81819f0f | 2261 | } |
2cfb7455 | 2262 | |
f6e7def7 | 2263 | page = c->page; |
497b66f2 | 2264 | if (likely(!kmem_cache_debug(s))) |
4b6f0750 | 2265 | goto load_freelist; |
2cfb7455 | 2266 | |
497b66f2 | 2267 | /* Only entered in the debug case */ |
f6e7def7 | 2268 | if (!alloc_debug_processing(s, page, freelist, addr)) |
497b66f2 | 2269 | goto new_slab; /* Slab failed checks. Next slab needed */ |
894b8788 | 2270 | |
f6e7def7 | 2271 | deactivate_slab(s, page, get_freepointer(s, freelist)); |
c17dda40 CL |
2272 | c->page = NULL; |
2273 | c->freelist = NULL; | |
a71ae47a | 2274 | local_irq_restore(flags); |
6faa6833 | 2275 | return freelist; |
894b8788 CL |
2276 | } |
2277 | ||
2278 | /* | |
2279 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
2280 | * have the fastpath folded into their functions. So no function call | |
2281 | * overhead for requests that can be satisfied on the fastpath. | |
2282 | * | |
2283 | * The fastpath works by first checking if the lockless freelist can be used. | |
2284 | * If not then __slab_alloc is called for slow processing. | |
2285 | * | |
2286 | * Otherwise we can simply pick the next object from the lockless free list. | |
2287 | */ | |
06428780 | 2288 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
ce71e27c | 2289 | gfp_t gfpflags, int node, unsigned long addr) |
894b8788 | 2290 | { |
894b8788 | 2291 | void **object; |
dfb4f096 | 2292 | struct kmem_cache_cpu *c; |
57d437d2 | 2293 | struct page *page; |
8a5ec0ba | 2294 | unsigned long tid; |
1f84260c | 2295 | |
c016b0bd | 2296 | if (slab_pre_alloc_hook(s, gfpflags)) |
773ff60e | 2297 | return NULL; |
1f84260c | 2298 | |
8a5ec0ba | 2299 | redo: |
8a5ec0ba CL |
2300 | |
2301 | /* | |
2302 | * Must read kmem_cache cpu data via this cpu ptr. Preemption is | |
2303 | * enabled. We may switch back and forth between cpus while | |
2304 | * reading from one cpu area. That does not matter as long | |
2305 | * as we end up on the original cpu again when doing the cmpxchg. | |
2306 | */ | |
9dfc6e68 | 2307 | c = __this_cpu_ptr(s->cpu_slab); |
8a5ec0ba | 2308 | |
8a5ec0ba CL |
2309 | /* |
2310 | * The transaction ids are globally unique per cpu and per operation on | |
2311 | * a per cpu queue. Thus they can be guarantee that the cmpxchg_double | |
2312 | * occurs on the right processor and that there was no operation on the | |
2313 | * linked list in between. | |
2314 | */ | |
2315 | tid = c->tid; | |
2316 | barrier(); | |
8a5ec0ba | 2317 | |
9dfc6e68 | 2318 | object = c->freelist; |
57d437d2 CL |
2319 | page = c->page; |
2320 | if (unlikely(!object || !node_match(page, node))) | |
894b8788 | 2321 | |
dfb4f096 | 2322 | object = __slab_alloc(s, gfpflags, node, addr, c); |
894b8788 CL |
2323 | |
2324 | else { | |
0ad9500e ED |
2325 | void *next_object = get_freepointer_safe(s, object); |
2326 | ||
8a5ec0ba | 2327 | /* |
25985edc | 2328 | * The cmpxchg will only match if there was no additional |
8a5ec0ba CL |
2329 | * operation and if we are on the right processor. |
2330 | * | |
2331 | * The cmpxchg does the following atomically (without lock semantics!) | |
2332 | * 1. Relocate first pointer to the current per cpu area. | |
2333 | * 2. Verify that tid and freelist have not been changed | |
2334 | * 3. If they were not changed replace tid and freelist | |
2335 | * | |
2336 | * Since this is without lock semantics the protection is only against | |
2337 | * code executing on this cpu *not* from access by other cpus. | |
2338 | */ | |
933393f5 | 2339 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2340 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2341 | object, tid, | |
0ad9500e | 2342 | next_object, next_tid(tid)))) { |
8a5ec0ba CL |
2343 | |
2344 | note_cmpxchg_failure("slab_alloc", s, tid); | |
2345 | goto redo; | |
2346 | } | |
0ad9500e | 2347 | prefetch_freepointer(s, next_object); |
84e554e6 | 2348 | stat(s, ALLOC_FASTPATH); |
894b8788 | 2349 | } |
8a5ec0ba | 2350 | |
74e2134f | 2351 | if (unlikely(gfpflags & __GFP_ZERO) && object) |
3b0efdfa | 2352 | memset(object, 0, s->object_size); |
d07dbea4 | 2353 | |
c016b0bd | 2354 | slab_post_alloc_hook(s, gfpflags, object); |
5a896d9e | 2355 | |
894b8788 | 2356 | return object; |
81819f0f CL |
2357 | } |
2358 | ||
2359 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) | |
2360 | { | |
2154a336 | 2361 | void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_); |
5b882be4 | 2362 | |
3b0efdfa | 2363 | trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size, s->size, gfpflags); |
5b882be4 EGM |
2364 | |
2365 | return ret; | |
81819f0f CL |
2366 | } |
2367 | EXPORT_SYMBOL(kmem_cache_alloc); | |
2368 | ||
0f24f128 | 2369 | #ifdef CONFIG_TRACING |
4a92379b RK |
2370 | void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size) |
2371 | { | |
2372 | void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_); | |
2373 | trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags); | |
2374 | return ret; | |
2375 | } | |
2376 | EXPORT_SYMBOL(kmem_cache_alloc_trace); | |
2377 | ||
2378 | void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) | |
5b882be4 | 2379 | { |
4a92379b RK |
2380 | void *ret = kmalloc_order(size, flags, order); |
2381 | trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags); | |
2382 | return ret; | |
5b882be4 | 2383 | } |
4a92379b | 2384 | EXPORT_SYMBOL(kmalloc_order_trace); |
5b882be4 EGM |
2385 | #endif |
2386 | ||
81819f0f CL |
2387 | #ifdef CONFIG_NUMA |
2388 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
2389 | { | |
5b882be4 EGM |
2390 | void *ret = slab_alloc(s, gfpflags, node, _RET_IP_); |
2391 | ||
ca2b84cb | 2392 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
3b0efdfa | 2393 | s->object_size, s->size, gfpflags, node); |
5b882be4 EGM |
2394 | |
2395 | return ret; | |
81819f0f CL |
2396 | } |
2397 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
81819f0f | 2398 | |
0f24f128 | 2399 | #ifdef CONFIG_TRACING |
4a92379b | 2400 | void *kmem_cache_alloc_node_trace(struct kmem_cache *s, |
5b882be4 | 2401 | gfp_t gfpflags, |
4a92379b | 2402 | int node, size_t size) |
5b882be4 | 2403 | { |
4a92379b RK |
2404 | void *ret = slab_alloc(s, gfpflags, node, _RET_IP_); |
2405 | ||
2406 | trace_kmalloc_node(_RET_IP_, ret, | |
2407 | size, s->size, gfpflags, node); | |
2408 | return ret; | |
5b882be4 | 2409 | } |
4a92379b | 2410 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
5b882be4 | 2411 | #endif |
5d1f57e4 | 2412 | #endif |
5b882be4 | 2413 | |
81819f0f | 2414 | /* |
894b8788 CL |
2415 | * Slow patch handling. This may still be called frequently since objects |
2416 | * have a longer lifetime than the cpu slabs in most processing loads. | |
81819f0f | 2417 | * |
894b8788 CL |
2418 | * So we still attempt to reduce cache line usage. Just take the slab |
2419 | * lock and free the item. If there is no additional partial page | |
2420 | * handling required then we can return immediately. | |
81819f0f | 2421 | */ |
894b8788 | 2422 | static void __slab_free(struct kmem_cache *s, struct page *page, |
ff12059e | 2423 | void *x, unsigned long addr) |
81819f0f CL |
2424 | { |
2425 | void *prior; | |
2426 | void **object = (void *)x; | |
2cfb7455 CL |
2427 | int was_frozen; |
2428 | int inuse; | |
2429 | struct page new; | |
2430 | unsigned long counters; | |
2431 | struct kmem_cache_node *n = NULL; | |
61728d1e | 2432 | unsigned long uninitialized_var(flags); |
81819f0f | 2433 | |
8a5ec0ba | 2434 | stat(s, FREE_SLOWPATH); |
81819f0f | 2435 | |
8dc16c6c | 2436 | if (kmem_cache_debug(s) && !free_debug_processing(s, page, x, addr)) |
80f08c19 | 2437 | return; |
6446faa2 | 2438 | |
2cfb7455 CL |
2439 | do { |
2440 | prior = page->freelist; | |
2441 | counters = page->counters; | |
2442 | set_freepointer(s, object, prior); | |
2443 | new.counters = counters; | |
2444 | was_frozen = new.frozen; | |
2445 | new.inuse--; | |
2446 | if ((!new.inuse || !prior) && !was_frozen && !n) { | |
49e22585 CL |
2447 | |
2448 | if (!kmem_cache_debug(s) && !prior) | |
2449 | ||
2450 | /* | |
2451 | * Slab was on no list before and will be partially empty | |
2452 | * We can defer the list move and instead freeze it. | |
2453 | */ | |
2454 | new.frozen = 1; | |
2455 | ||
2456 | else { /* Needs to be taken off a list */ | |
2457 | ||
2458 | n = get_node(s, page_to_nid(page)); | |
2459 | /* | |
2460 | * Speculatively acquire the list_lock. | |
2461 | * If the cmpxchg does not succeed then we may | |
2462 | * drop the list_lock without any processing. | |
2463 | * | |
2464 | * Otherwise the list_lock will synchronize with | |
2465 | * other processors updating the list of slabs. | |
2466 | */ | |
2467 | spin_lock_irqsave(&n->list_lock, flags); | |
2468 | ||
2469 | } | |
2cfb7455 CL |
2470 | } |
2471 | inuse = new.inuse; | |
81819f0f | 2472 | |
2cfb7455 CL |
2473 | } while (!cmpxchg_double_slab(s, page, |
2474 | prior, counters, | |
2475 | object, new.counters, | |
2476 | "__slab_free")); | |
81819f0f | 2477 | |
2cfb7455 | 2478 | if (likely(!n)) { |
49e22585 CL |
2479 | |
2480 | /* | |
2481 | * If we just froze the page then put it onto the | |
2482 | * per cpu partial list. | |
2483 | */ | |
8028dcea | 2484 | if (new.frozen && !was_frozen) { |
49e22585 | 2485 | put_cpu_partial(s, page, 1); |
8028dcea AS |
2486 | stat(s, CPU_PARTIAL_FREE); |
2487 | } | |
49e22585 | 2488 | /* |
2cfb7455 CL |
2489 | * The list lock was not taken therefore no list |
2490 | * activity can be necessary. | |
2491 | */ | |
2492 | if (was_frozen) | |
2493 | stat(s, FREE_FROZEN); | |
80f08c19 | 2494 | return; |
2cfb7455 | 2495 | } |
81819f0f CL |
2496 | |
2497 | /* | |
2cfb7455 CL |
2498 | * was_frozen may have been set after we acquired the list_lock in |
2499 | * an earlier loop. So we need to check it here again. | |
81819f0f | 2500 | */ |
2cfb7455 CL |
2501 | if (was_frozen) |
2502 | stat(s, FREE_FROZEN); | |
2503 | else { | |
2504 | if (unlikely(!inuse && n->nr_partial > s->min_partial)) | |
2505 | goto slab_empty; | |
81819f0f | 2506 | |
2cfb7455 CL |
2507 | /* |
2508 | * Objects left in the slab. If it was not on the partial list before | |
2509 | * then add it. | |
2510 | */ | |
2511 | if (unlikely(!prior)) { | |
2512 | remove_full(s, page); | |
136333d1 | 2513 | add_partial(n, page, DEACTIVATE_TO_TAIL); |
2cfb7455 CL |
2514 | stat(s, FREE_ADD_PARTIAL); |
2515 | } | |
8ff12cfc | 2516 | } |
80f08c19 | 2517 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
2518 | return; |
2519 | ||
2520 | slab_empty: | |
a973e9dd | 2521 | if (prior) { |
81819f0f | 2522 | /* |
6fbabb20 | 2523 | * Slab on the partial list. |
81819f0f | 2524 | */ |
5cc6eee8 | 2525 | remove_partial(n, page); |
84e554e6 | 2526 | stat(s, FREE_REMOVE_PARTIAL); |
6fbabb20 CL |
2527 | } else |
2528 | /* Slab must be on the full list */ | |
2529 | remove_full(s, page); | |
2cfb7455 | 2530 | |
80f08c19 | 2531 | spin_unlock_irqrestore(&n->list_lock, flags); |
84e554e6 | 2532 | stat(s, FREE_SLAB); |
81819f0f | 2533 | discard_slab(s, page); |
81819f0f CL |
2534 | } |
2535 | ||
894b8788 CL |
2536 | /* |
2537 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
2538 | * can perform fastpath freeing without additional function calls. | |
2539 | * | |
2540 | * The fastpath is only possible if we are freeing to the current cpu slab | |
2541 | * of this processor. This typically the case if we have just allocated | |
2542 | * the item before. | |
2543 | * | |
2544 | * If fastpath is not possible then fall back to __slab_free where we deal | |
2545 | * with all sorts of special processing. | |
2546 | */ | |
06428780 | 2547 | static __always_inline void slab_free(struct kmem_cache *s, |
ce71e27c | 2548 | struct page *page, void *x, unsigned long addr) |
894b8788 CL |
2549 | { |
2550 | void **object = (void *)x; | |
dfb4f096 | 2551 | struct kmem_cache_cpu *c; |
8a5ec0ba | 2552 | unsigned long tid; |
1f84260c | 2553 | |
c016b0bd CL |
2554 | slab_free_hook(s, x); |
2555 | ||
8a5ec0ba CL |
2556 | redo: |
2557 | /* | |
2558 | * Determine the currently cpus per cpu slab. | |
2559 | * The cpu may change afterward. However that does not matter since | |
2560 | * data is retrieved via this pointer. If we are on the same cpu | |
2561 | * during the cmpxchg then the free will succedd. | |
2562 | */ | |
9dfc6e68 | 2563 | c = __this_cpu_ptr(s->cpu_slab); |
c016b0bd | 2564 | |
8a5ec0ba CL |
2565 | tid = c->tid; |
2566 | barrier(); | |
c016b0bd | 2567 | |
442b06bc | 2568 | if (likely(page == c->page)) { |
ff12059e | 2569 | set_freepointer(s, object, c->freelist); |
8a5ec0ba | 2570 | |
933393f5 | 2571 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2572 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2573 | c->freelist, tid, | |
2574 | object, next_tid(tid)))) { | |
2575 | ||
2576 | note_cmpxchg_failure("slab_free", s, tid); | |
2577 | goto redo; | |
2578 | } | |
84e554e6 | 2579 | stat(s, FREE_FASTPATH); |
894b8788 | 2580 | } else |
ff12059e | 2581 | __slab_free(s, page, x, addr); |
894b8788 | 2582 | |
894b8788 CL |
2583 | } |
2584 | ||
81819f0f CL |
2585 | void kmem_cache_free(struct kmem_cache *s, void *x) |
2586 | { | |
77c5e2d0 | 2587 | struct page *page; |
81819f0f | 2588 | |
b49af68f | 2589 | page = virt_to_head_page(x); |
81819f0f | 2590 | |
ce71e27c | 2591 | slab_free(s, page, x, _RET_IP_); |
5b882be4 | 2592 | |
ca2b84cb | 2593 | trace_kmem_cache_free(_RET_IP_, x); |
81819f0f CL |
2594 | } |
2595 | EXPORT_SYMBOL(kmem_cache_free); | |
2596 | ||
81819f0f | 2597 | /* |
672bba3a CL |
2598 | * Object placement in a slab is made very easy because we always start at |
2599 | * offset 0. If we tune the size of the object to the alignment then we can | |
2600 | * get the required alignment by putting one properly sized object after | |
2601 | * another. | |
81819f0f CL |
2602 | * |
2603 | * Notice that the allocation order determines the sizes of the per cpu | |
2604 | * caches. Each processor has always one slab available for allocations. | |
2605 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 2606 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 2607 | * locking overhead. |
81819f0f CL |
2608 | */ |
2609 | ||
2610 | /* | |
2611 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
2612 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
2613 | * and increases the number of allocations possible without having to | |
2614 | * take the list_lock. | |
2615 | */ | |
2616 | static int slub_min_order; | |
114e9e89 | 2617 | static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; |
9b2cd506 | 2618 | static int slub_min_objects; |
81819f0f CL |
2619 | |
2620 | /* | |
2621 | * Merge control. If this is set then no merging of slab caches will occur. | |
672bba3a | 2622 | * (Could be removed. This was introduced to pacify the merge skeptics.) |
81819f0f CL |
2623 | */ |
2624 | static int slub_nomerge; | |
2625 | ||
81819f0f CL |
2626 | /* |
2627 | * Calculate the order of allocation given an slab object size. | |
2628 | * | |
672bba3a CL |
2629 | * The order of allocation has significant impact on performance and other |
2630 | * system components. Generally order 0 allocations should be preferred since | |
2631 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
2632 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 2633 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
2634 | * would be wasted. |
2635 | * | |
2636 | * In order to reach satisfactory performance we must ensure that a minimum | |
2637 | * number of objects is in one slab. Otherwise we may generate too much | |
2638 | * activity on the partial lists which requires taking the list_lock. This is | |
2639 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 2640 | * |
672bba3a CL |
2641 | * slub_max_order specifies the order where we begin to stop considering the |
2642 | * number of objects in a slab as critical. If we reach slub_max_order then | |
2643 | * we try to keep the page order as low as possible. So we accept more waste | |
2644 | * of space in favor of a small page order. | |
81819f0f | 2645 | * |
672bba3a CL |
2646 | * Higher order allocations also allow the placement of more objects in a |
2647 | * slab and thereby reduce object handling overhead. If the user has | |
2648 | * requested a higher mininum order then we start with that one instead of | |
2649 | * the smallest order which will fit the object. | |
81819f0f | 2650 | */ |
5e6d444e | 2651 | static inline int slab_order(int size, int min_objects, |
ab9a0f19 | 2652 | int max_order, int fract_leftover, int reserved) |
81819f0f CL |
2653 | { |
2654 | int order; | |
2655 | int rem; | |
6300ea75 | 2656 | int min_order = slub_min_order; |
81819f0f | 2657 | |
ab9a0f19 | 2658 | if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE) |
210b5c06 | 2659 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; |
39b26464 | 2660 | |
6300ea75 | 2661 | for (order = max(min_order, |
5e6d444e CL |
2662 | fls(min_objects * size - 1) - PAGE_SHIFT); |
2663 | order <= max_order; order++) { | |
81819f0f | 2664 | |
5e6d444e | 2665 | unsigned long slab_size = PAGE_SIZE << order; |
81819f0f | 2666 | |
ab9a0f19 | 2667 | if (slab_size < min_objects * size + reserved) |
81819f0f CL |
2668 | continue; |
2669 | ||
ab9a0f19 | 2670 | rem = (slab_size - reserved) % size; |
81819f0f | 2671 | |
5e6d444e | 2672 | if (rem <= slab_size / fract_leftover) |
81819f0f CL |
2673 | break; |
2674 | ||
2675 | } | |
672bba3a | 2676 | |
81819f0f CL |
2677 | return order; |
2678 | } | |
2679 | ||
ab9a0f19 | 2680 | static inline int calculate_order(int size, int reserved) |
5e6d444e CL |
2681 | { |
2682 | int order; | |
2683 | int min_objects; | |
2684 | int fraction; | |
e8120ff1 | 2685 | int max_objects; |
5e6d444e CL |
2686 | |
2687 | /* | |
2688 | * Attempt to find best configuration for a slab. This | |
2689 | * works by first attempting to generate a layout with | |
2690 | * the best configuration and backing off gradually. | |
2691 | * | |
2692 | * First we reduce the acceptable waste in a slab. Then | |
2693 | * we reduce the minimum objects required in a slab. | |
2694 | */ | |
2695 | min_objects = slub_min_objects; | |
9b2cd506 CL |
2696 | if (!min_objects) |
2697 | min_objects = 4 * (fls(nr_cpu_ids) + 1); | |
ab9a0f19 | 2698 | max_objects = order_objects(slub_max_order, size, reserved); |
e8120ff1 ZY |
2699 | min_objects = min(min_objects, max_objects); |
2700 | ||
5e6d444e | 2701 | while (min_objects > 1) { |
c124f5b5 | 2702 | fraction = 16; |
5e6d444e CL |
2703 | while (fraction >= 4) { |
2704 | order = slab_order(size, min_objects, | |
ab9a0f19 | 2705 | slub_max_order, fraction, reserved); |
5e6d444e CL |
2706 | if (order <= slub_max_order) |
2707 | return order; | |
2708 | fraction /= 2; | |
2709 | } | |
5086c389 | 2710 | min_objects--; |
5e6d444e CL |
2711 | } |
2712 | ||
2713 | /* | |
2714 | * We were unable to place multiple objects in a slab. Now | |
2715 | * lets see if we can place a single object there. | |
2716 | */ | |
ab9a0f19 | 2717 | order = slab_order(size, 1, slub_max_order, 1, reserved); |
5e6d444e CL |
2718 | if (order <= slub_max_order) |
2719 | return order; | |
2720 | ||
2721 | /* | |
2722 | * Doh this slab cannot be placed using slub_max_order. | |
2723 | */ | |
ab9a0f19 | 2724 | order = slab_order(size, 1, MAX_ORDER, 1, reserved); |
818cf590 | 2725 | if (order < MAX_ORDER) |
5e6d444e CL |
2726 | return order; |
2727 | return -ENOSYS; | |
2728 | } | |
2729 | ||
81819f0f | 2730 | /* |
672bba3a | 2731 | * Figure out what the alignment of the objects will be. |
81819f0f CL |
2732 | */ |
2733 | static unsigned long calculate_alignment(unsigned long flags, | |
2734 | unsigned long align, unsigned long size) | |
2735 | { | |
2736 | /* | |
6446faa2 CL |
2737 | * If the user wants hardware cache aligned objects then follow that |
2738 | * suggestion if the object is sufficiently large. | |
81819f0f | 2739 | * |
6446faa2 CL |
2740 | * The hardware cache alignment cannot override the specified |
2741 | * alignment though. If that is greater then use it. | |
81819f0f | 2742 | */ |
b6210386 NP |
2743 | if (flags & SLAB_HWCACHE_ALIGN) { |
2744 | unsigned long ralign = cache_line_size(); | |
2745 | while (size <= ralign / 2) | |
2746 | ralign /= 2; | |
2747 | align = max(align, ralign); | |
2748 | } | |
81819f0f CL |
2749 | |
2750 | if (align < ARCH_SLAB_MINALIGN) | |
b6210386 | 2751 | align = ARCH_SLAB_MINALIGN; |
81819f0f CL |
2752 | |
2753 | return ALIGN(align, sizeof(void *)); | |
2754 | } | |
2755 | ||
5595cffc | 2756 | static void |
4053497d | 2757 | init_kmem_cache_node(struct kmem_cache_node *n) |
81819f0f CL |
2758 | { |
2759 | n->nr_partial = 0; | |
81819f0f CL |
2760 | spin_lock_init(&n->list_lock); |
2761 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 2762 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 2763 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 2764 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 2765 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 2766 | #endif |
81819f0f CL |
2767 | } |
2768 | ||
55136592 | 2769 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s) |
4c93c355 | 2770 | { |
6c182dc0 CL |
2771 | BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE < |
2772 | SLUB_PAGE_SHIFT * sizeof(struct kmem_cache_cpu)); | |
4c93c355 | 2773 | |
8a5ec0ba | 2774 | /* |
d4d84fef CM |
2775 | * Must align to double word boundary for the double cmpxchg |
2776 | * instructions to work; see __pcpu_double_call_return_bool(). | |
8a5ec0ba | 2777 | */ |
d4d84fef CM |
2778 | s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), |
2779 | 2 * sizeof(void *)); | |
8a5ec0ba CL |
2780 | |
2781 | if (!s->cpu_slab) | |
2782 | return 0; | |
2783 | ||
2784 | init_kmem_cache_cpus(s); | |
4c93c355 | 2785 | |
8a5ec0ba | 2786 | return 1; |
4c93c355 | 2787 | } |
4c93c355 | 2788 | |
51df1142 CL |
2789 | static struct kmem_cache *kmem_cache_node; |
2790 | ||
81819f0f CL |
2791 | /* |
2792 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
2793 | * slab on the node for this slabcache. There are no concurrent accesses | |
2794 | * possible. | |
2795 | * | |
2796 | * Note that this function only works on the kmalloc_node_cache | |
4c93c355 CL |
2797 | * when allocating for the kmalloc_node_cache. This is used for bootstrapping |
2798 | * memory on a fresh node that has no slab structures yet. | |
81819f0f | 2799 | */ |
55136592 | 2800 | static void early_kmem_cache_node_alloc(int node) |
81819f0f CL |
2801 | { |
2802 | struct page *page; | |
2803 | struct kmem_cache_node *n; | |
2804 | ||
51df1142 | 2805 | BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node)); |
81819f0f | 2806 | |
51df1142 | 2807 | page = new_slab(kmem_cache_node, GFP_NOWAIT, node); |
81819f0f CL |
2808 | |
2809 | BUG_ON(!page); | |
a2f92ee7 CL |
2810 | if (page_to_nid(page) != node) { |
2811 | printk(KERN_ERR "SLUB: Unable to allocate memory from " | |
2812 | "node %d\n", node); | |
2813 | printk(KERN_ERR "SLUB: Allocating a useless per node structure " | |
2814 | "in order to be able to continue\n"); | |
2815 | } | |
2816 | ||
81819f0f CL |
2817 | n = page->freelist; |
2818 | BUG_ON(!n); | |
51df1142 | 2819 | page->freelist = get_freepointer(kmem_cache_node, n); |
e6e82ea1 | 2820 | page->inuse = 1; |
8cb0a506 | 2821 | page->frozen = 0; |
51df1142 | 2822 | kmem_cache_node->node[node] = n; |
8ab1372f | 2823 | #ifdef CONFIG_SLUB_DEBUG |
f7cb1933 | 2824 | init_object(kmem_cache_node, n, SLUB_RED_ACTIVE); |
51df1142 | 2825 | init_tracking(kmem_cache_node, n); |
8ab1372f | 2826 | #endif |
4053497d | 2827 | init_kmem_cache_node(n); |
51df1142 | 2828 | inc_slabs_node(kmem_cache_node, node, page->objects); |
6446faa2 | 2829 | |
136333d1 | 2830 | add_partial(n, page, DEACTIVATE_TO_HEAD); |
81819f0f CL |
2831 | } |
2832 | ||
2833 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
2834 | { | |
2835 | int node; | |
2836 | ||
f64dc58c | 2837 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f | 2838 | struct kmem_cache_node *n = s->node[node]; |
51df1142 | 2839 | |
73367bd8 | 2840 | if (n) |
51df1142 CL |
2841 | kmem_cache_free(kmem_cache_node, n); |
2842 | ||
81819f0f CL |
2843 | s->node[node] = NULL; |
2844 | } | |
2845 | } | |
2846 | ||
55136592 | 2847 | static int init_kmem_cache_nodes(struct kmem_cache *s) |
81819f0f CL |
2848 | { |
2849 | int node; | |
81819f0f | 2850 | |
f64dc58c | 2851 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2852 | struct kmem_cache_node *n; |
2853 | ||
73367bd8 | 2854 | if (slab_state == DOWN) { |
55136592 | 2855 | early_kmem_cache_node_alloc(node); |
73367bd8 AD |
2856 | continue; |
2857 | } | |
51df1142 | 2858 | n = kmem_cache_alloc_node(kmem_cache_node, |
55136592 | 2859 | GFP_KERNEL, node); |
81819f0f | 2860 | |
73367bd8 AD |
2861 | if (!n) { |
2862 | free_kmem_cache_nodes(s); | |
2863 | return 0; | |
81819f0f | 2864 | } |
73367bd8 | 2865 | |
81819f0f | 2866 | s->node[node] = n; |
4053497d | 2867 | init_kmem_cache_node(n); |
81819f0f CL |
2868 | } |
2869 | return 1; | |
2870 | } | |
81819f0f | 2871 | |
c0bdb232 | 2872 | static void set_min_partial(struct kmem_cache *s, unsigned long min) |
3b89d7d8 DR |
2873 | { |
2874 | if (min < MIN_PARTIAL) | |
2875 | min = MIN_PARTIAL; | |
2876 | else if (min > MAX_PARTIAL) | |
2877 | min = MAX_PARTIAL; | |
2878 | s->min_partial = min; | |
2879 | } | |
2880 | ||
81819f0f CL |
2881 | /* |
2882 | * calculate_sizes() determines the order and the distribution of data within | |
2883 | * a slab object. | |
2884 | */ | |
06b285dc | 2885 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f CL |
2886 | { |
2887 | unsigned long flags = s->flags; | |
3b0efdfa | 2888 | unsigned long size = s->object_size; |
81819f0f | 2889 | unsigned long align = s->align; |
834f3d11 | 2890 | int order; |
81819f0f | 2891 | |
d8b42bf5 CL |
2892 | /* |
2893 | * Round up object size to the next word boundary. We can only | |
2894 | * place the free pointer at word boundaries and this determines | |
2895 | * the possible location of the free pointer. | |
2896 | */ | |
2897 | size = ALIGN(size, sizeof(void *)); | |
2898 | ||
2899 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
2900 | /* |
2901 | * Determine if we can poison the object itself. If the user of | |
2902 | * the slab may touch the object after free or before allocation | |
2903 | * then we should never poison the object itself. | |
2904 | */ | |
2905 | if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) && | |
c59def9f | 2906 | !s->ctor) |
81819f0f CL |
2907 | s->flags |= __OBJECT_POISON; |
2908 | else | |
2909 | s->flags &= ~__OBJECT_POISON; | |
2910 | ||
81819f0f CL |
2911 | |
2912 | /* | |
672bba3a | 2913 | * If we are Redzoning then check if there is some space between the |
81819f0f | 2914 | * end of the object and the free pointer. If not then add an |
672bba3a | 2915 | * additional word to have some bytes to store Redzone information. |
81819f0f | 2916 | */ |
3b0efdfa | 2917 | if ((flags & SLAB_RED_ZONE) && size == s->object_size) |
81819f0f | 2918 | size += sizeof(void *); |
41ecc55b | 2919 | #endif |
81819f0f CL |
2920 | |
2921 | /* | |
672bba3a CL |
2922 | * With that we have determined the number of bytes in actual use |
2923 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
2924 | */ |
2925 | s->inuse = size; | |
2926 | ||
2927 | if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) || | |
c59def9f | 2928 | s->ctor)) { |
81819f0f CL |
2929 | /* |
2930 | * Relocate free pointer after the object if it is not | |
2931 | * permitted to overwrite the first word of the object on | |
2932 | * kmem_cache_free. | |
2933 | * | |
2934 | * This is the case if we do RCU, have a constructor or | |
2935 | * destructor or are poisoning the objects. | |
2936 | */ | |
2937 | s->offset = size; | |
2938 | size += sizeof(void *); | |
2939 | } | |
2940 | ||
c12b3c62 | 2941 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
2942 | if (flags & SLAB_STORE_USER) |
2943 | /* | |
2944 | * Need to store information about allocs and frees after | |
2945 | * the object. | |
2946 | */ | |
2947 | size += 2 * sizeof(struct track); | |
2948 | ||
be7b3fbc | 2949 | if (flags & SLAB_RED_ZONE) |
81819f0f CL |
2950 | /* |
2951 | * Add some empty padding so that we can catch | |
2952 | * overwrites from earlier objects rather than let | |
2953 | * tracking information or the free pointer be | |
0211a9c8 | 2954 | * corrupted if a user writes before the start |
81819f0f CL |
2955 | * of the object. |
2956 | */ | |
2957 | size += sizeof(void *); | |
41ecc55b | 2958 | #endif |
672bba3a | 2959 | |
81819f0f CL |
2960 | /* |
2961 | * Determine the alignment based on various parameters that the | |
65c02d4c CL |
2962 | * user specified and the dynamic determination of cache line size |
2963 | * on bootup. | |
81819f0f | 2964 | */ |
3b0efdfa | 2965 | align = calculate_alignment(flags, align, s->object_size); |
dcb0ce1b | 2966 | s->align = align; |
81819f0f CL |
2967 | |
2968 | /* | |
2969 | * SLUB stores one object immediately after another beginning from | |
2970 | * offset 0. In order to align the objects we have to simply size | |
2971 | * each object to conform to the alignment. | |
2972 | */ | |
2973 | size = ALIGN(size, align); | |
2974 | s->size = size; | |
06b285dc CL |
2975 | if (forced_order >= 0) |
2976 | order = forced_order; | |
2977 | else | |
ab9a0f19 | 2978 | order = calculate_order(size, s->reserved); |
81819f0f | 2979 | |
834f3d11 | 2980 | if (order < 0) |
81819f0f CL |
2981 | return 0; |
2982 | ||
b7a49f0d | 2983 | s->allocflags = 0; |
834f3d11 | 2984 | if (order) |
b7a49f0d CL |
2985 | s->allocflags |= __GFP_COMP; |
2986 | ||
2987 | if (s->flags & SLAB_CACHE_DMA) | |
2988 | s->allocflags |= SLUB_DMA; | |
2989 | ||
2990 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
2991 | s->allocflags |= __GFP_RECLAIMABLE; | |
2992 | ||
81819f0f CL |
2993 | /* |
2994 | * Determine the number of objects per slab | |
2995 | */ | |
ab9a0f19 LJ |
2996 | s->oo = oo_make(order, size, s->reserved); |
2997 | s->min = oo_make(get_order(size), size, s->reserved); | |
205ab99d CL |
2998 | if (oo_objects(s->oo) > oo_objects(s->max)) |
2999 | s->max = s->oo; | |
81819f0f | 3000 | |
834f3d11 | 3001 | return !!oo_objects(s->oo); |
81819f0f CL |
3002 | |
3003 | } | |
3004 | ||
55136592 | 3005 | static int kmem_cache_open(struct kmem_cache *s, |
81819f0f CL |
3006 | const char *name, size_t size, |
3007 | size_t align, unsigned long flags, | |
51cc5068 | 3008 | void (*ctor)(void *)) |
81819f0f CL |
3009 | { |
3010 | memset(s, 0, kmem_size); | |
3011 | s->name = name; | |
3012 | s->ctor = ctor; | |
3b0efdfa | 3013 | s->object_size = size; |
81819f0f | 3014 | s->align = align; |
ba0268a8 | 3015 | s->flags = kmem_cache_flags(size, flags, name, ctor); |
ab9a0f19 | 3016 | s->reserved = 0; |
81819f0f | 3017 | |
da9a638c LJ |
3018 | if (need_reserve_slab_rcu && (s->flags & SLAB_DESTROY_BY_RCU)) |
3019 | s->reserved = sizeof(struct rcu_head); | |
81819f0f | 3020 | |
06b285dc | 3021 | if (!calculate_sizes(s, -1)) |
81819f0f | 3022 | goto error; |
3de47213 DR |
3023 | if (disable_higher_order_debug) { |
3024 | /* | |
3025 | * Disable debugging flags that store metadata if the min slab | |
3026 | * order increased. | |
3027 | */ | |
3b0efdfa | 3028 | if (get_order(s->size) > get_order(s->object_size)) { |
3de47213 DR |
3029 | s->flags &= ~DEBUG_METADATA_FLAGS; |
3030 | s->offset = 0; | |
3031 | if (!calculate_sizes(s, -1)) | |
3032 | goto error; | |
3033 | } | |
3034 | } | |
81819f0f | 3035 | |
2565409f HC |
3036 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
3037 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
b789ef51 CL |
3038 | if (system_has_cmpxchg_double() && (s->flags & SLAB_DEBUG_FLAGS) == 0) |
3039 | /* Enable fast mode */ | |
3040 | s->flags |= __CMPXCHG_DOUBLE; | |
3041 | #endif | |
3042 | ||
3b89d7d8 DR |
3043 | /* |
3044 | * The larger the object size is, the more pages we want on the partial | |
3045 | * list to avoid pounding the page allocator excessively. | |
3046 | */ | |
49e22585 CL |
3047 | set_min_partial(s, ilog2(s->size) / 2); |
3048 | ||
3049 | /* | |
3050 | * cpu_partial determined the maximum number of objects kept in the | |
3051 | * per cpu partial lists of a processor. | |
3052 | * | |
3053 | * Per cpu partial lists mainly contain slabs that just have one | |
3054 | * object freed. If they are used for allocation then they can be | |
3055 | * filled up again with minimal effort. The slab will never hit the | |
3056 | * per node partial lists and therefore no locking will be required. | |
3057 | * | |
3058 | * This setting also determines | |
3059 | * | |
3060 | * A) The number of objects from per cpu partial slabs dumped to the | |
3061 | * per node list when we reach the limit. | |
9f264904 | 3062 | * B) The number of objects in cpu partial slabs to extract from the |
49e22585 CL |
3063 | * per node list when we run out of per cpu objects. We only fetch 50% |
3064 | * to keep some capacity around for frees. | |
3065 | */ | |
8f1e33da CL |
3066 | if (kmem_cache_debug(s)) |
3067 | s->cpu_partial = 0; | |
3068 | else if (s->size >= PAGE_SIZE) | |
49e22585 CL |
3069 | s->cpu_partial = 2; |
3070 | else if (s->size >= 1024) | |
3071 | s->cpu_partial = 6; | |
3072 | else if (s->size >= 256) | |
3073 | s->cpu_partial = 13; | |
3074 | else | |
3075 | s->cpu_partial = 30; | |
3076 | ||
81819f0f CL |
3077 | s->refcount = 1; |
3078 | #ifdef CONFIG_NUMA | |
e2cb96b7 | 3079 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 3080 | #endif |
55136592 | 3081 | if (!init_kmem_cache_nodes(s)) |
dfb4f096 | 3082 | goto error; |
81819f0f | 3083 | |
55136592 | 3084 | if (alloc_kmem_cache_cpus(s)) |
81819f0f | 3085 | return 1; |
ff12059e | 3086 | |
4c93c355 | 3087 | free_kmem_cache_nodes(s); |
81819f0f CL |
3088 | error: |
3089 | if (flags & SLAB_PANIC) | |
3090 | panic("Cannot create slab %s size=%lu realsize=%u " | |
3091 | "order=%u offset=%u flags=%lx\n", | |
834f3d11 | 3092 | s->name, (unsigned long)size, s->size, oo_order(s->oo), |
81819f0f CL |
3093 | s->offset, flags); |
3094 | return 0; | |
3095 | } | |
81819f0f | 3096 | |
81819f0f CL |
3097 | /* |
3098 | * Determine the size of a slab object | |
3099 | */ | |
3100 | unsigned int kmem_cache_size(struct kmem_cache *s) | |
3101 | { | |
3b0efdfa | 3102 | return s->object_size; |
81819f0f CL |
3103 | } |
3104 | EXPORT_SYMBOL(kmem_cache_size); | |
3105 | ||
33b12c38 CL |
3106 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
3107 | const char *text) | |
3108 | { | |
3109 | #ifdef CONFIG_SLUB_DEBUG | |
3110 | void *addr = page_address(page); | |
3111 | void *p; | |
a5dd5c11 NK |
3112 | unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) * |
3113 | sizeof(long), GFP_ATOMIC); | |
bbd7d57b ED |
3114 | if (!map) |
3115 | return; | |
33b12c38 CL |
3116 | slab_err(s, page, "%s", text); |
3117 | slab_lock(page); | |
33b12c38 | 3118 | |
5f80b13a | 3119 | get_map(s, page, map); |
33b12c38 CL |
3120 | for_each_object(p, s, addr, page->objects) { |
3121 | ||
3122 | if (!test_bit(slab_index(p, s, addr), map)) { | |
3123 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n", | |
3124 | p, p - addr); | |
3125 | print_tracking(s, p); | |
3126 | } | |
3127 | } | |
3128 | slab_unlock(page); | |
bbd7d57b | 3129 | kfree(map); |
33b12c38 CL |
3130 | #endif |
3131 | } | |
3132 | ||
81819f0f | 3133 | /* |
599870b1 | 3134 | * Attempt to free all partial slabs on a node. |
69cb8e6b CL |
3135 | * This is called from kmem_cache_close(). We must be the last thread |
3136 | * using the cache and therefore we do not need to lock anymore. | |
81819f0f | 3137 | */ |
599870b1 | 3138 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 3139 | { |
81819f0f CL |
3140 | struct page *page, *h; |
3141 | ||
33b12c38 | 3142 | list_for_each_entry_safe(page, h, &n->partial, lru) { |
81819f0f | 3143 | if (!page->inuse) { |
5cc6eee8 | 3144 | remove_partial(n, page); |
81819f0f | 3145 | discard_slab(s, page); |
33b12c38 CL |
3146 | } else { |
3147 | list_slab_objects(s, page, | |
3148 | "Objects remaining on kmem_cache_close()"); | |
599870b1 | 3149 | } |
33b12c38 | 3150 | } |
81819f0f CL |
3151 | } |
3152 | ||
3153 | /* | |
672bba3a | 3154 | * Release all resources used by a slab cache. |
81819f0f | 3155 | */ |
0c710013 | 3156 | static inline int kmem_cache_close(struct kmem_cache *s) |
81819f0f CL |
3157 | { |
3158 | int node; | |
3159 | ||
3160 | flush_all(s); | |
9dfc6e68 | 3161 | free_percpu(s->cpu_slab); |
81819f0f | 3162 | /* Attempt to free all objects */ |
f64dc58c | 3163 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
3164 | struct kmem_cache_node *n = get_node(s, node); |
3165 | ||
599870b1 CL |
3166 | free_partial(s, n); |
3167 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
3168 | return 1; |
3169 | } | |
3170 | free_kmem_cache_nodes(s); | |
3171 | return 0; | |
3172 | } | |
3173 | ||
3174 | /* | |
3175 | * Close a cache and release the kmem_cache structure | |
3176 | * (must be used for caches created using kmem_cache_create) | |
3177 | */ | |
3178 | void kmem_cache_destroy(struct kmem_cache *s) | |
3179 | { | |
3180 | down_write(&slub_lock); | |
3181 | s->refcount--; | |
3182 | if (!s->refcount) { | |
3183 | list_del(&s->list); | |
69cb8e6b | 3184 | up_write(&slub_lock); |
d629d819 PE |
3185 | if (kmem_cache_close(s)) { |
3186 | printk(KERN_ERR "SLUB %s: %s called for cache that " | |
3187 | "still has objects.\n", s->name, __func__); | |
3188 | dump_stack(); | |
3189 | } | |
d76b1590 ED |
3190 | if (s->flags & SLAB_DESTROY_BY_RCU) |
3191 | rcu_barrier(); | |
81819f0f | 3192 | sysfs_slab_remove(s); |
69cb8e6b CL |
3193 | } else |
3194 | up_write(&slub_lock); | |
81819f0f CL |
3195 | } |
3196 | EXPORT_SYMBOL(kmem_cache_destroy); | |
3197 | ||
3198 | /******************************************************************** | |
3199 | * Kmalloc subsystem | |
3200 | *******************************************************************/ | |
3201 | ||
51df1142 | 3202 | struct kmem_cache *kmalloc_caches[SLUB_PAGE_SHIFT]; |
81819f0f CL |
3203 | EXPORT_SYMBOL(kmalloc_caches); |
3204 | ||
51df1142 CL |
3205 | static struct kmem_cache *kmem_cache; |
3206 | ||
55136592 | 3207 | #ifdef CONFIG_ZONE_DMA |
51df1142 | 3208 | static struct kmem_cache *kmalloc_dma_caches[SLUB_PAGE_SHIFT]; |
55136592 CL |
3209 | #endif |
3210 | ||
81819f0f CL |
3211 | static int __init setup_slub_min_order(char *str) |
3212 | { | |
06428780 | 3213 | get_option(&str, &slub_min_order); |
81819f0f CL |
3214 | |
3215 | return 1; | |
3216 | } | |
3217 | ||
3218 | __setup("slub_min_order=", setup_slub_min_order); | |
3219 | ||
3220 | static int __init setup_slub_max_order(char *str) | |
3221 | { | |
06428780 | 3222 | get_option(&str, &slub_max_order); |
818cf590 | 3223 | slub_max_order = min(slub_max_order, MAX_ORDER - 1); |
81819f0f CL |
3224 | |
3225 | return 1; | |
3226 | } | |
3227 | ||
3228 | __setup("slub_max_order=", setup_slub_max_order); | |
3229 | ||
3230 | static int __init setup_slub_min_objects(char *str) | |
3231 | { | |
06428780 | 3232 | get_option(&str, &slub_min_objects); |
81819f0f CL |
3233 | |
3234 | return 1; | |
3235 | } | |
3236 | ||
3237 | __setup("slub_min_objects=", setup_slub_min_objects); | |
3238 | ||
3239 | static int __init setup_slub_nomerge(char *str) | |
3240 | { | |
3241 | slub_nomerge = 1; | |
3242 | return 1; | |
3243 | } | |
3244 | ||
3245 | __setup("slub_nomerge", setup_slub_nomerge); | |
3246 | ||
51df1142 CL |
3247 | static struct kmem_cache *__init create_kmalloc_cache(const char *name, |
3248 | int size, unsigned int flags) | |
81819f0f | 3249 | { |
51df1142 CL |
3250 | struct kmem_cache *s; |
3251 | ||
3252 | s = kmem_cache_alloc(kmem_cache, GFP_NOWAIT); | |
3253 | ||
83b519e8 PE |
3254 | /* |
3255 | * This function is called with IRQs disabled during early-boot on | |
3256 | * single CPU so there's no need to take slub_lock here. | |
3257 | */ | |
55136592 | 3258 | if (!kmem_cache_open(s, name, size, ARCH_KMALLOC_MINALIGN, |
319d1e24 | 3259 | flags, NULL)) |
81819f0f CL |
3260 | goto panic; |
3261 | ||
3262 | list_add(&s->list, &slab_caches); | |
51df1142 | 3263 | return s; |
81819f0f CL |
3264 | |
3265 | panic: | |
3266 | panic("Creation of kmalloc slab %s size=%d failed.\n", name, size); | |
51df1142 | 3267 | return NULL; |
81819f0f CL |
3268 | } |
3269 | ||
f1b26339 CL |
3270 | /* |
3271 | * Conversion table for small slabs sizes / 8 to the index in the | |
3272 | * kmalloc array. This is necessary for slabs < 192 since we have non power | |
3273 | * of two cache sizes there. The size of larger slabs can be determined using | |
3274 | * fls. | |
3275 | */ | |
3276 | static s8 size_index[24] = { | |
3277 | 3, /* 8 */ | |
3278 | 4, /* 16 */ | |
3279 | 5, /* 24 */ | |
3280 | 5, /* 32 */ | |
3281 | 6, /* 40 */ | |
3282 | 6, /* 48 */ | |
3283 | 6, /* 56 */ | |
3284 | 6, /* 64 */ | |
3285 | 1, /* 72 */ | |
3286 | 1, /* 80 */ | |
3287 | 1, /* 88 */ | |
3288 | 1, /* 96 */ | |
3289 | 7, /* 104 */ | |
3290 | 7, /* 112 */ | |
3291 | 7, /* 120 */ | |
3292 | 7, /* 128 */ | |
3293 | 2, /* 136 */ | |
3294 | 2, /* 144 */ | |
3295 | 2, /* 152 */ | |
3296 | 2, /* 160 */ | |
3297 | 2, /* 168 */ | |
3298 | 2, /* 176 */ | |
3299 | 2, /* 184 */ | |
3300 | 2 /* 192 */ | |
3301 | }; | |
3302 | ||
acdfcd04 AK |
3303 | static inline int size_index_elem(size_t bytes) |
3304 | { | |
3305 | return (bytes - 1) / 8; | |
3306 | } | |
3307 | ||
81819f0f CL |
3308 | static struct kmem_cache *get_slab(size_t size, gfp_t flags) |
3309 | { | |
f1b26339 | 3310 | int index; |
81819f0f | 3311 | |
f1b26339 CL |
3312 | if (size <= 192) { |
3313 | if (!size) | |
3314 | return ZERO_SIZE_PTR; | |
81819f0f | 3315 | |
acdfcd04 | 3316 | index = size_index[size_index_elem(size)]; |
aadb4bc4 | 3317 | } else |
f1b26339 | 3318 | index = fls(size - 1); |
81819f0f CL |
3319 | |
3320 | #ifdef CONFIG_ZONE_DMA | |
f1b26339 | 3321 | if (unlikely((flags & SLUB_DMA))) |
51df1142 | 3322 | return kmalloc_dma_caches[index]; |
f1b26339 | 3323 | |
81819f0f | 3324 | #endif |
51df1142 | 3325 | return kmalloc_caches[index]; |
81819f0f CL |
3326 | } |
3327 | ||
3328 | void *__kmalloc(size_t size, gfp_t flags) | |
3329 | { | |
aadb4bc4 | 3330 | struct kmem_cache *s; |
5b882be4 | 3331 | void *ret; |
81819f0f | 3332 | |
ffadd4d0 | 3333 | if (unlikely(size > SLUB_MAX_SIZE)) |
eada35ef | 3334 | return kmalloc_large(size, flags); |
aadb4bc4 CL |
3335 | |
3336 | s = get_slab(size, flags); | |
3337 | ||
3338 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3339 | return s; |
3340 | ||
2154a336 | 3341 | ret = slab_alloc(s, flags, NUMA_NO_NODE, _RET_IP_); |
5b882be4 | 3342 | |
ca2b84cb | 3343 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 EGM |
3344 | |
3345 | return ret; | |
81819f0f CL |
3346 | } |
3347 | EXPORT_SYMBOL(__kmalloc); | |
3348 | ||
5d1f57e4 | 3349 | #ifdef CONFIG_NUMA |
f619cfe1 CL |
3350 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
3351 | { | |
b1eeab67 | 3352 | struct page *page; |
e4f7c0b4 | 3353 | void *ptr = NULL; |
f619cfe1 | 3354 | |
b1eeab67 VN |
3355 | flags |= __GFP_COMP | __GFP_NOTRACK; |
3356 | page = alloc_pages_node(node, flags, get_order(size)); | |
f619cfe1 | 3357 | if (page) |
e4f7c0b4 CM |
3358 | ptr = page_address(page); |
3359 | ||
3360 | kmemleak_alloc(ptr, size, 1, flags); | |
3361 | return ptr; | |
f619cfe1 CL |
3362 | } |
3363 | ||
81819f0f CL |
3364 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3365 | { | |
aadb4bc4 | 3366 | struct kmem_cache *s; |
5b882be4 | 3367 | void *ret; |
81819f0f | 3368 | |
057685cf | 3369 | if (unlikely(size > SLUB_MAX_SIZE)) { |
5b882be4 EGM |
3370 | ret = kmalloc_large_node(size, flags, node); |
3371 | ||
ca2b84cb EGM |
3372 | trace_kmalloc_node(_RET_IP_, ret, |
3373 | size, PAGE_SIZE << get_order(size), | |
3374 | flags, node); | |
5b882be4 EGM |
3375 | |
3376 | return ret; | |
3377 | } | |
aadb4bc4 CL |
3378 | |
3379 | s = get_slab(size, flags); | |
3380 | ||
3381 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3382 | return s; |
3383 | ||
5b882be4 EGM |
3384 | ret = slab_alloc(s, flags, node, _RET_IP_); |
3385 | ||
ca2b84cb | 3386 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 EGM |
3387 | |
3388 | return ret; | |
81819f0f CL |
3389 | } |
3390 | EXPORT_SYMBOL(__kmalloc_node); | |
3391 | #endif | |
3392 | ||
3393 | size_t ksize(const void *object) | |
3394 | { | |
272c1d21 | 3395 | struct page *page; |
81819f0f | 3396 | |
ef8b4520 | 3397 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
3398 | return 0; |
3399 | ||
294a80a8 | 3400 | page = virt_to_head_page(object); |
294a80a8 | 3401 | |
76994412 PE |
3402 | if (unlikely(!PageSlab(page))) { |
3403 | WARN_ON(!PageCompound(page)); | |
294a80a8 | 3404 | return PAGE_SIZE << compound_order(page); |
76994412 | 3405 | } |
81819f0f | 3406 | |
b3d41885 | 3407 | return slab_ksize(page->slab); |
81819f0f | 3408 | } |
b1aabecd | 3409 | EXPORT_SYMBOL(ksize); |
81819f0f | 3410 | |
d18a90dd BG |
3411 | #ifdef CONFIG_SLUB_DEBUG |
3412 | bool verify_mem_not_deleted(const void *x) | |
3413 | { | |
3414 | struct page *page; | |
3415 | void *object = (void *)x; | |
3416 | unsigned long flags; | |
3417 | bool rv; | |
3418 | ||
3419 | if (unlikely(ZERO_OR_NULL_PTR(x))) | |
3420 | return false; | |
3421 | ||
3422 | local_irq_save(flags); | |
3423 | ||
3424 | page = virt_to_head_page(x); | |
3425 | if (unlikely(!PageSlab(page))) { | |
3426 | /* maybe it was from stack? */ | |
3427 | rv = true; | |
3428 | goto out_unlock; | |
3429 | } | |
3430 | ||
3431 | slab_lock(page); | |
3432 | if (on_freelist(page->slab, page, object)) { | |
3433 | object_err(page->slab, page, object, "Object is on free-list"); | |
3434 | rv = false; | |
3435 | } else { | |
3436 | rv = true; | |
3437 | } | |
3438 | slab_unlock(page); | |
3439 | ||
3440 | out_unlock: | |
3441 | local_irq_restore(flags); | |
3442 | return rv; | |
3443 | } | |
3444 | EXPORT_SYMBOL(verify_mem_not_deleted); | |
3445 | #endif | |
3446 | ||
81819f0f CL |
3447 | void kfree(const void *x) |
3448 | { | |
81819f0f | 3449 | struct page *page; |
5bb983b0 | 3450 | void *object = (void *)x; |
81819f0f | 3451 | |
2121db74 PE |
3452 | trace_kfree(_RET_IP_, x); |
3453 | ||
2408c550 | 3454 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
3455 | return; |
3456 | ||
b49af68f | 3457 | page = virt_to_head_page(x); |
aadb4bc4 | 3458 | if (unlikely(!PageSlab(page))) { |
0937502a | 3459 | BUG_ON(!PageCompound(page)); |
e4f7c0b4 | 3460 | kmemleak_free(x); |
aadb4bc4 CL |
3461 | put_page(page); |
3462 | return; | |
3463 | } | |
ce71e27c | 3464 | slab_free(page->slab, page, object, _RET_IP_); |
81819f0f CL |
3465 | } |
3466 | EXPORT_SYMBOL(kfree); | |
3467 | ||
2086d26a | 3468 | /* |
672bba3a CL |
3469 | * kmem_cache_shrink removes empty slabs from the partial lists and sorts |
3470 | * the remaining slabs by the number of items in use. The slabs with the | |
3471 | * most items in use come first. New allocations will then fill those up | |
3472 | * and thus they can be removed from the partial lists. | |
3473 | * | |
3474 | * The slabs with the least items are placed last. This results in them | |
3475 | * being allocated from last increasing the chance that the last objects | |
3476 | * are freed in them. | |
2086d26a CL |
3477 | */ |
3478 | int kmem_cache_shrink(struct kmem_cache *s) | |
3479 | { | |
3480 | int node; | |
3481 | int i; | |
3482 | struct kmem_cache_node *n; | |
3483 | struct page *page; | |
3484 | struct page *t; | |
205ab99d | 3485 | int objects = oo_objects(s->max); |
2086d26a | 3486 | struct list_head *slabs_by_inuse = |
834f3d11 | 3487 | kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL); |
2086d26a CL |
3488 | unsigned long flags; |
3489 | ||
3490 | if (!slabs_by_inuse) | |
3491 | return -ENOMEM; | |
3492 | ||
3493 | flush_all(s); | |
f64dc58c | 3494 | for_each_node_state(node, N_NORMAL_MEMORY) { |
2086d26a CL |
3495 | n = get_node(s, node); |
3496 | ||
3497 | if (!n->nr_partial) | |
3498 | continue; | |
3499 | ||
834f3d11 | 3500 | for (i = 0; i < objects; i++) |
2086d26a CL |
3501 | INIT_LIST_HEAD(slabs_by_inuse + i); |
3502 | ||
3503 | spin_lock_irqsave(&n->list_lock, flags); | |
3504 | ||
3505 | /* | |
672bba3a | 3506 | * Build lists indexed by the items in use in each slab. |
2086d26a | 3507 | * |
672bba3a CL |
3508 | * Note that concurrent frees may occur while we hold the |
3509 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
3510 | */ |
3511 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
69cb8e6b CL |
3512 | list_move(&page->lru, slabs_by_inuse + page->inuse); |
3513 | if (!page->inuse) | |
3514 | n->nr_partial--; | |
2086d26a CL |
3515 | } |
3516 | ||
2086d26a | 3517 | /* |
672bba3a CL |
3518 | * Rebuild the partial list with the slabs filled up most |
3519 | * first and the least used slabs at the end. | |
2086d26a | 3520 | */ |
69cb8e6b | 3521 | for (i = objects - 1; i > 0; i--) |
2086d26a CL |
3522 | list_splice(slabs_by_inuse + i, n->partial.prev); |
3523 | ||
2086d26a | 3524 | spin_unlock_irqrestore(&n->list_lock, flags); |
69cb8e6b CL |
3525 | |
3526 | /* Release empty slabs */ | |
3527 | list_for_each_entry_safe(page, t, slabs_by_inuse, lru) | |
3528 | discard_slab(s, page); | |
2086d26a CL |
3529 | } |
3530 | ||
3531 | kfree(slabs_by_inuse); | |
3532 | return 0; | |
3533 | } | |
3534 | EXPORT_SYMBOL(kmem_cache_shrink); | |
3535 | ||
92a5bbc1 | 3536 | #if defined(CONFIG_MEMORY_HOTPLUG) |
b9049e23 YG |
3537 | static int slab_mem_going_offline_callback(void *arg) |
3538 | { | |
3539 | struct kmem_cache *s; | |
3540 | ||
3541 | down_read(&slub_lock); | |
3542 | list_for_each_entry(s, &slab_caches, list) | |
3543 | kmem_cache_shrink(s); | |
3544 | up_read(&slub_lock); | |
3545 | ||
3546 | return 0; | |
3547 | } | |
3548 | ||
3549 | static void slab_mem_offline_callback(void *arg) | |
3550 | { | |
3551 | struct kmem_cache_node *n; | |
3552 | struct kmem_cache *s; | |
3553 | struct memory_notify *marg = arg; | |
3554 | int offline_node; | |
3555 | ||
3556 | offline_node = marg->status_change_nid; | |
3557 | ||
3558 | /* | |
3559 | * If the node still has available memory. we need kmem_cache_node | |
3560 | * for it yet. | |
3561 | */ | |
3562 | if (offline_node < 0) | |
3563 | return; | |
3564 | ||
3565 | down_read(&slub_lock); | |
3566 | list_for_each_entry(s, &slab_caches, list) { | |
3567 | n = get_node(s, offline_node); | |
3568 | if (n) { | |
3569 | /* | |
3570 | * if n->nr_slabs > 0, slabs still exist on the node | |
3571 | * that is going down. We were unable to free them, | |
c9404c9c | 3572 | * and offline_pages() function shouldn't call this |
b9049e23 YG |
3573 | * callback. So, we must fail. |
3574 | */ | |
0f389ec6 | 3575 | BUG_ON(slabs_node(s, offline_node)); |
b9049e23 YG |
3576 | |
3577 | s->node[offline_node] = NULL; | |
8de66a0c | 3578 | kmem_cache_free(kmem_cache_node, n); |
b9049e23 YG |
3579 | } |
3580 | } | |
3581 | up_read(&slub_lock); | |
3582 | } | |
3583 | ||
3584 | static int slab_mem_going_online_callback(void *arg) | |
3585 | { | |
3586 | struct kmem_cache_node *n; | |
3587 | struct kmem_cache *s; | |
3588 | struct memory_notify *marg = arg; | |
3589 | int nid = marg->status_change_nid; | |
3590 | int ret = 0; | |
3591 | ||
3592 | /* | |
3593 | * If the node's memory is already available, then kmem_cache_node is | |
3594 | * already created. Nothing to do. | |
3595 | */ | |
3596 | if (nid < 0) | |
3597 | return 0; | |
3598 | ||
3599 | /* | |
0121c619 | 3600 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
3601 | * allocate a kmem_cache_node structure in order to bring the node |
3602 | * online. | |
3603 | */ | |
3604 | down_read(&slub_lock); | |
3605 | list_for_each_entry(s, &slab_caches, list) { | |
3606 | /* | |
3607 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
3608 | * since memory is not yet available from the node that | |
3609 | * is brought up. | |
3610 | */ | |
8de66a0c | 3611 | n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL); |
b9049e23 YG |
3612 | if (!n) { |
3613 | ret = -ENOMEM; | |
3614 | goto out; | |
3615 | } | |
4053497d | 3616 | init_kmem_cache_node(n); |
b9049e23 YG |
3617 | s->node[nid] = n; |
3618 | } | |
3619 | out: | |
3620 | up_read(&slub_lock); | |
3621 | return ret; | |
3622 | } | |
3623 | ||
3624 | static int slab_memory_callback(struct notifier_block *self, | |
3625 | unsigned long action, void *arg) | |
3626 | { | |
3627 | int ret = 0; | |
3628 | ||
3629 | switch (action) { | |
3630 | case MEM_GOING_ONLINE: | |
3631 | ret = slab_mem_going_online_callback(arg); | |
3632 | break; | |
3633 | case MEM_GOING_OFFLINE: | |
3634 | ret = slab_mem_going_offline_callback(arg); | |
3635 | break; | |
3636 | case MEM_OFFLINE: | |
3637 | case MEM_CANCEL_ONLINE: | |
3638 | slab_mem_offline_callback(arg); | |
3639 | break; | |
3640 | case MEM_ONLINE: | |
3641 | case MEM_CANCEL_OFFLINE: | |
3642 | break; | |
3643 | } | |
dc19f9db KH |
3644 | if (ret) |
3645 | ret = notifier_from_errno(ret); | |
3646 | else | |
3647 | ret = NOTIFY_OK; | |
b9049e23 YG |
3648 | return ret; |
3649 | } | |
3650 | ||
3651 | #endif /* CONFIG_MEMORY_HOTPLUG */ | |
3652 | ||
81819f0f CL |
3653 | /******************************************************************** |
3654 | * Basic setup of slabs | |
3655 | *******************************************************************/ | |
3656 | ||
51df1142 CL |
3657 | /* |
3658 | * Used for early kmem_cache structures that were allocated using | |
3659 | * the page allocator | |
3660 | */ | |
3661 | ||
3662 | static void __init kmem_cache_bootstrap_fixup(struct kmem_cache *s) | |
3663 | { | |
3664 | int node; | |
3665 | ||
3666 | list_add(&s->list, &slab_caches); | |
3667 | s->refcount = -1; | |
3668 | ||
3669 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
3670 | struct kmem_cache_node *n = get_node(s, node); | |
3671 | struct page *p; | |
3672 | ||
3673 | if (n) { | |
3674 | list_for_each_entry(p, &n->partial, lru) | |
3675 | p->slab = s; | |
3676 | ||
607bf324 | 3677 | #ifdef CONFIG_SLUB_DEBUG |
51df1142 CL |
3678 | list_for_each_entry(p, &n->full, lru) |
3679 | p->slab = s; | |
3680 | #endif | |
3681 | } | |
3682 | } | |
3683 | } | |
3684 | ||
81819f0f CL |
3685 | void __init kmem_cache_init(void) |
3686 | { | |
3687 | int i; | |
4b356be0 | 3688 | int caches = 0; |
51df1142 CL |
3689 | struct kmem_cache *temp_kmem_cache; |
3690 | int order; | |
51df1142 CL |
3691 | struct kmem_cache *temp_kmem_cache_node; |
3692 | unsigned long kmalloc_size; | |
3693 | ||
fc8d8620 SG |
3694 | if (debug_guardpage_minorder()) |
3695 | slub_max_order = 0; | |
3696 | ||
51df1142 CL |
3697 | kmem_size = offsetof(struct kmem_cache, node) + |
3698 | nr_node_ids * sizeof(struct kmem_cache_node *); | |
3699 | ||
3700 | /* Allocate two kmem_caches from the page allocator */ | |
3701 | kmalloc_size = ALIGN(kmem_size, cache_line_size()); | |
3702 | order = get_order(2 * kmalloc_size); | |
3703 | kmem_cache = (void *)__get_free_pages(GFP_NOWAIT, order); | |
3704 | ||
81819f0f CL |
3705 | /* |
3706 | * Must first have the slab cache available for the allocations of the | |
672bba3a | 3707 | * struct kmem_cache_node's. There is special bootstrap code in |
81819f0f CL |
3708 | * kmem_cache_open for slab_state == DOWN. |
3709 | */ | |
51df1142 CL |
3710 | kmem_cache_node = (void *)kmem_cache + kmalloc_size; |
3711 | ||
3712 | kmem_cache_open(kmem_cache_node, "kmem_cache_node", | |
3713 | sizeof(struct kmem_cache_node), | |
3714 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); | |
b9049e23 | 3715 | |
0c40ba4f | 3716 | hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI); |
81819f0f CL |
3717 | |
3718 | /* Able to allocate the per node structures */ | |
3719 | slab_state = PARTIAL; | |
3720 | ||
51df1142 CL |
3721 | temp_kmem_cache = kmem_cache; |
3722 | kmem_cache_open(kmem_cache, "kmem_cache", kmem_size, | |
3723 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); | |
3724 | kmem_cache = kmem_cache_alloc(kmem_cache, GFP_NOWAIT); | |
3725 | memcpy(kmem_cache, temp_kmem_cache, kmem_size); | |
81819f0f | 3726 | |
51df1142 CL |
3727 | /* |
3728 | * Allocate kmem_cache_node properly from the kmem_cache slab. | |
3729 | * kmem_cache_node is separately allocated so no need to | |
3730 | * update any list pointers. | |
3731 | */ | |
3732 | temp_kmem_cache_node = kmem_cache_node; | |
81819f0f | 3733 | |
51df1142 CL |
3734 | kmem_cache_node = kmem_cache_alloc(kmem_cache, GFP_NOWAIT); |
3735 | memcpy(kmem_cache_node, temp_kmem_cache_node, kmem_size); | |
3736 | ||
3737 | kmem_cache_bootstrap_fixup(kmem_cache_node); | |
3738 | ||
3739 | caches++; | |
51df1142 CL |
3740 | kmem_cache_bootstrap_fixup(kmem_cache); |
3741 | caches++; | |
3742 | /* Free temporary boot structure */ | |
3743 | free_pages((unsigned long)temp_kmem_cache, order); | |
3744 | ||
3745 | /* Now we can use the kmem_cache to allocate kmalloc slabs */ | |
f1b26339 CL |
3746 | |
3747 | /* | |
3748 | * Patch up the size_index table if we have strange large alignment | |
3749 | * requirements for the kmalloc array. This is only the case for | |
6446faa2 | 3750 | * MIPS it seems. The standard arches will not generate any code here. |
f1b26339 CL |
3751 | * |
3752 | * Largest permitted alignment is 256 bytes due to the way we | |
3753 | * handle the index determination for the smaller caches. | |
3754 | * | |
3755 | * Make sure that nothing crazy happens if someone starts tinkering | |
3756 | * around with ARCH_KMALLOC_MINALIGN | |
3757 | */ | |
3758 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || | |
3759 | (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1))); | |
3760 | ||
acdfcd04 AK |
3761 | for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) { |
3762 | int elem = size_index_elem(i); | |
3763 | if (elem >= ARRAY_SIZE(size_index)) | |
3764 | break; | |
3765 | size_index[elem] = KMALLOC_SHIFT_LOW; | |
3766 | } | |
f1b26339 | 3767 | |
acdfcd04 AK |
3768 | if (KMALLOC_MIN_SIZE == 64) { |
3769 | /* | |
3770 | * The 96 byte size cache is not used if the alignment | |
3771 | * is 64 byte. | |
3772 | */ | |
3773 | for (i = 64 + 8; i <= 96; i += 8) | |
3774 | size_index[size_index_elem(i)] = 7; | |
3775 | } else if (KMALLOC_MIN_SIZE == 128) { | |
41d54d3b CL |
3776 | /* |
3777 | * The 192 byte sized cache is not used if the alignment | |
3778 | * is 128 byte. Redirect kmalloc to use the 256 byte cache | |
3779 | * instead. | |
3780 | */ | |
3781 | for (i = 128 + 8; i <= 192; i += 8) | |
acdfcd04 | 3782 | size_index[size_index_elem(i)] = 8; |
41d54d3b CL |
3783 | } |
3784 | ||
51df1142 CL |
3785 | /* Caches that are not of the two-to-the-power-of size */ |
3786 | if (KMALLOC_MIN_SIZE <= 32) { | |
3787 | kmalloc_caches[1] = create_kmalloc_cache("kmalloc-96", 96, 0); | |
3788 | caches++; | |
3789 | } | |
3790 | ||
3791 | if (KMALLOC_MIN_SIZE <= 64) { | |
3792 | kmalloc_caches[2] = create_kmalloc_cache("kmalloc-192", 192, 0); | |
3793 | caches++; | |
3794 | } | |
3795 | ||
3796 | for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) { | |
3797 | kmalloc_caches[i] = create_kmalloc_cache("kmalloc", 1 << i, 0); | |
3798 | caches++; | |
3799 | } | |
3800 | ||
81819f0f CL |
3801 | slab_state = UP; |
3802 | ||
3803 | /* Provide the correct kmalloc names now that the caches are up */ | |
84c1cf62 PE |
3804 | if (KMALLOC_MIN_SIZE <= 32) { |
3805 | kmalloc_caches[1]->name = kstrdup(kmalloc_caches[1]->name, GFP_NOWAIT); | |
3806 | BUG_ON(!kmalloc_caches[1]->name); | |
3807 | } | |
3808 | ||
3809 | if (KMALLOC_MIN_SIZE <= 64) { | |
3810 | kmalloc_caches[2]->name = kstrdup(kmalloc_caches[2]->name, GFP_NOWAIT); | |
3811 | BUG_ON(!kmalloc_caches[2]->name); | |
3812 | } | |
3813 | ||
d7278bd7 CL |
3814 | for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) { |
3815 | char *s = kasprintf(GFP_NOWAIT, "kmalloc-%d", 1 << i); | |
3816 | ||
3817 | BUG_ON(!s); | |
51df1142 | 3818 | kmalloc_caches[i]->name = s; |
d7278bd7 | 3819 | } |
81819f0f CL |
3820 | |
3821 | #ifdef CONFIG_SMP | |
3822 | register_cpu_notifier(&slab_notifier); | |
9dfc6e68 | 3823 | #endif |
81819f0f | 3824 | |
55136592 | 3825 | #ifdef CONFIG_ZONE_DMA |
51df1142 CL |
3826 | for (i = 0; i < SLUB_PAGE_SHIFT; i++) { |
3827 | struct kmem_cache *s = kmalloc_caches[i]; | |
55136592 | 3828 | |
51df1142 | 3829 | if (s && s->size) { |
55136592 | 3830 | char *name = kasprintf(GFP_NOWAIT, |
3b0efdfa | 3831 | "dma-kmalloc-%d", s->object_size); |
55136592 CL |
3832 | |
3833 | BUG_ON(!name); | |
51df1142 | 3834 | kmalloc_dma_caches[i] = create_kmalloc_cache(name, |
3b0efdfa | 3835 | s->object_size, SLAB_CACHE_DMA); |
55136592 CL |
3836 | } |
3837 | } | |
3838 | #endif | |
3adbefee IM |
3839 | printk(KERN_INFO |
3840 | "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d," | |
4b356be0 CL |
3841 | " CPUs=%d, Nodes=%d\n", |
3842 | caches, cache_line_size(), | |
81819f0f CL |
3843 | slub_min_order, slub_max_order, slub_min_objects, |
3844 | nr_cpu_ids, nr_node_ids); | |
3845 | } | |
3846 | ||
7e85ee0c PE |
3847 | void __init kmem_cache_init_late(void) |
3848 | { | |
7e85ee0c PE |
3849 | } |
3850 | ||
81819f0f CL |
3851 | /* |
3852 | * Find a mergeable slab cache | |
3853 | */ | |
3854 | static int slab_unmergeable(struct kmem_cache *s) | |
3855 | { | |
3856 | if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE)) | |
3857 | return 1; | |
3858 | ||
c59def9f | 3859 | if (s->ctor) |
81819f0f CL |
3860 | return 1; |
3861 | ||
8ffa6875 CL |
3862 | /* |
3863 | * We may have set a slab to be unmergeable during bootstrap. | |
3864 | */ | |
3865 | if (s->refcount < 0) | |
3866 | return 1; | |
3867 | ||
81819f0f CL |
3868 | return 0; |
3869 | } | |
3870 | ||
3871 | static struct kmem_cache *find_mergeable(size_t size, | |
ba0268a8 | 3872 | size_t align, unsigned long flags, const char *name, |
51cc5068 | 3873 | void (*ctor)(void *)) |
81819f0f | 3874 | { |
5b95a4ac | 3875 | struct kmem_cache *s; |
81819f0f CL |
3876 | |
3877 | if (slub_nomerge || (flags & SLUB_NEVER_MERGE)) | |
3878 | return NULL; | |
3879 | ||
c59def9f | 3880 | if (ctor) |
81819f0f CL |
3881 | return NULL; |
3882 | ||
3883 | size = ALIGN(size, sizeof(void *)); | |
3884 | align = calculate_alignment(flags, align, size); | |
3885 | size = ALIGN(size, align); | |
ba0268a8 | 3886 | flags = kmem_cache_flags(size, flags, name, NULL); |
81819f0f | 3887 | |
5b95a4ac | 3888 | list_for_each_entry(s, &slab_caches, list) { |
81819f0f CL |
3889 | if (slab_unmergeable(s)) |
3890 | continue; | |
3891 | ||
3892 | if (size > s->size) | |
3893 | continue; | |
3894 | ||
ba0268a8 | 3895 | if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME)) |
81819f0f CL |
3896 | continue; |
3897 | /* | |
3898 | * Check if alignment is compatible. | |
3899 | * Courtesy of Adrian Drzewiecki | |
3900 | */ | |
06428780 | 3901 | if ((s->size & ~(align - 1)) != s->size) |
81819f0f CL |
3902 | continue; |
3903 | ||
3904 | if (s->size - size >= sizeof(void *)) | |
3905 | continue; | |
3906 | ||
3907 | return s; | |
3908 | } | |
3909 | return NULL; | |
3910 | } | |
3911 | ||
039363f3 | 3912 | struct kmem_cache *__kmem_cache_create(const char *name, size_t size, |
51cc5068 | 3913 | size_t align, unsigned long flags, void (*ctor)(void *)) |
81819f0f CL |
3914 | { |
3915 | struct kmem_cache *s; | |
84c1cf62 | 3916 | char *n; |
81819f0f CL |
3917 | |
3918 | down_write(&slub_lock); | |
ba0268a8 | 3919 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f CL |
3920 | if (s) { |
3921 | s->refcount++; | |
3922 | /* | |
3923 | * Adjust the object sizes so that we clear | |
3924 | * the complete object on kzalloc. | |
3925 | */ | |
3b0efdfa | 3926 | s->object_size = max(s->object_size, (int)size); |
81819f0f | 3927 | s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *))); |
6446faa2 | 3928 | |
7b8f3b66 | 3929 | if (sysfs_slab_alias(s, name)) { |
7b8f3b66 | 3930 | s->refcount--; |
81819f0f | 3931 | goto err; |
7b8f3b66 | 3932 | } |
2bce6485 | 3933 | up_write(&slub_lock); |
a0e1d1be CL |
3934 | return s; |
3935 | } | |
6446faa2 | 3936 | |
84c1cf62 PE |
3937 | n = kstrdup(name, GFP_KERNEL); |
3938 | if (!n) | |
3939 | goto err; | |
3940 | ||
a0e1d1be CL |
3941 | s = kmalloc(kmem_size, GFP_KERNEL); |
3942 | if (s) { | |
84c1cf62 | 3943 | if (kmem_cache_open(s, n, |
c59def9f | 3944 | size, align, flags, ctor)) { |
81819f0f | 3945 | list_add(&s->list, &slab_caches); |
66c4c35c | 3946 | up_write(&slub_lock); |
7b8f3b66 | 3947 | if (sysfs_slab_add(s)) { |
66c4c35c | 3948 | down_write(&slub_lock); |
7b8f3b66 | 3949 | list_del(&s->list); |
84c1cf62 | 3950 | kfree(n); |
7b8f3b66 | 3951 | kfree(s); |
a0e1d1be | 3952 | goto err; |
7b8f3b66 | 3953 | } |
a0e1d1be CL |
3954 | return s; |
3955 | } | |
3956 | kfree(s); | |
81819f0f | 3957 | } |
601d39d0 | 3958 | kfree(n); |
68cee4f1 | 3959 | err: |
81819f0f | 3960 | up_write(&slub_lock); |
81819f0f CL |
3961 | return s; |
3962 | } | |
81819f0f | 3963 | |
81819f0f | 3964 | #ifdef CONFIG_SMP |
81819f0f | 3965 | /* |
672bba3a CL |
3966 | * Use the cpu notifier to insure that the cpu slabs are flushed when |
3967 | * necessary. | |
81819f0f CL |
3968 | */ |
3969 | static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb, | |
3970 | unsigned long action, void *hcpu) | |
3971 | { | |
3972 | long cpu = (long)hcpu; | |
5b95a4ac CL |
3973 | struct kmem_cache *s; |
3974 | unsigned long flags; | |
81819f0f CL |
3975 | |
3976 | switch (action) { | |
3977 | case CPU_UP_CANCELED: | |
8bb78442 | 3978 | case CPU_UP_CANCELED_FROZEN: |
81819f0f | 3979 | case CPU_DEAD: |
8bb78442 | 3980 | case CPU_DEAD_FROZEN: |
5b95a4ac CL |
3981 | down_read(&slub_lock); |
3982 | list_for_each_entry(s, &slab_caches, list) { | |
3983 | local_irq_save(flags); | |
3984 | __flush_cpu_slab(s, cpu); | |
3985 | local_irq_restore(flags); | |
3986 | } | |
3987 | up_read(&slub_lock); | |
81819f0f CL |
3988 | break; |
3989 | default: | |
3990 | break; | |
3991 | } | |
3992 | return NOTIFY_OK; | |
3993 | } | |
3994 | ||
06428780 | 3995 | static struct notifier_block __cpuinitdata slab_notifier = { |
3adbefee | 3996 | .notifier_call = slab_cpuup_callback |
06428780 | 3997 | }; |
81819f0f CL |
3998 | |
3999 | #endif | |
4000 | ||
ce71e27c | 4001 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 4002 | { |
aadb4bc4 | 4003 | struct kmem_cache *s; |
94b528d0 | 4004 | void *ret; |
aadb4bc4 | 4005 | |
ffadd4d0 | 4006 | if (unlikely(size > SLUB_MAX_SIZE)) |
eada35ef PE |
4007 | return kmalloc_large(size, gfpflags); |
4008 | ||
aadb4bc4 | 4009 | s = get_slab(size, gfpflags); |
81819f0f | 4010 | |
2408c550 | 4011 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4012 | return s; |
81819f0f | 4013 | |
2154a336 | 4014 | ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, caller); |
94b528d0 | 4015 | |
25985edc | 4016 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4017 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
4018 | |
4019 | return ret; | |
81819f0f CL |
4020 | } |
4021 | ||
5d1f57e4 | 4022 | #ifdef CONFIG_NUMA |
81819f0f | 4023 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, |
ce71e27c | 4024 | int node, unsigned long caller) |
81819f0f | 4025 | { |
aadb4bc4 | 4026 | struct kmem_cache *s; |
94b528d0 | 4027 | void *ret; |
aadb4bc4 | 4028 | |
d3e14aa3 XF |
4029 | if (unlikely(size > SLUB_MAX_SIZE)) { |
4030 | ret = kmalloc_large_node(size, gfpflags, node); | |
4031 | ||
4032 | trace_kmalloc_node(caller, ret, | |
4033 | size, PAGE_SIZE << get_order(size), | |
4034 | gfpflags, node); | |
4035 | ||
4036 | return ret; | |
4037 | } | |
eada35ef | 4038 | |
aadb4bc4 | 4039 | s = get_slab(size, gfpflags); |
81819f0f | 4040 | |
2408c550 | 4041 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4042 | return s; |
81819f0f | 4043 | |
94b528d0 EGM |
4044 | ret = slab_alloc(s, gfpflags, node, caller); |
4045 | ||
25985edc | 4046 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4047 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
4048 | |
4049 | return ret; | |
81819f0f | 4050 | } |
5d1f57e4 | 4051 | #endif |
81819f0f | 4052 | |
ab4d5ed5 | 4053 | #ifdef CONFIG_SYSFS |
205ab99d CL |
4054 | static int count_inuse(struct page *page) |
4055 | { | |
4056 | return page->inuse; | |
4057 | } | |
4058 | ||
4059 | static int count_total(struct page *page) | |
4060 | { | |
4061 | return page->objects; | |
4062 | } | |
ab4d5ed5 | 4063 | #endif |
205ab99d | 4064 | |
ab4d5ed5 | 4065 | #ifdef CONFIG_SLUB_DEBUG |
434e245d CL |
4066 | static int validate_slab(struct kmem_cache *s, struct page *page, |
4067 | unsigned long *map) | |
53e15af0 CL |
4068 | { |
4069 | void *p; | |
a973e9dd | 4070 | void *addr = page_address(page); |
53e15af0 CL |
4071 | |
4072 | if (!check_slab(s, page) || | |
4073 | !on_freelist(s, page, NULL)) | |
4074 | return 0; | |
4075 | ||
4076 | /* Now we know that a valid freelist exists */ | |
39b26464 | 4077 | bitmap_zero(map, page->objects); |
53e15af0 | 4078 | |
5f80b13a CL |
4079 | get_map(s, page, map); |
4080 | for_each_object(p, s, addr, page->objects) { | |
4081 | if (test_bit(slab_index(p, s, addr), map)) | |
4082 | if (!check_object(s, page, p, SLUB_RED_INACTIVE)) | |
4083 | return 0; | |
53e15af0 CL |
4084 | } |
4085 | ||
224a88be | 4086 | for_each_object(p, s, addr, page->objects) |
7656c72b | 4087 | if (!test_bit(slab_index(p, s, addr), map)) |
37d57443 | 4088 | if (!check_object(s, page, p, SLUB_RED_ACTIVE)) |
53e15af0 CL |
4089 | return 0; |
4090 | return 1; | |
4091 | } | |
4092 | ||
434e245d CL |
4093 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, |
4094 | unsigned long *map) | |
53e15af0 | 4095 | { |
881db7fb CL |
4096 | slab_lock(page); |
4097 | validate_slab(s, page, map); | |
4098 | slab_unlock(page); | |
53e15af0 CL |
4099 | } |
4100 | ||
434e245d CL |
4101 | static int validate_slab_node(struct kmem_cache *s, |
4102 | struct kmem_cache_node *n, unsigned long *map) | |
53e15af0 CL |
4103 | { |
4104 | unsigned long count = 0; | |
4105 | struct page *page; | |
4106 | unsigned long flags; | |
4107 | ||
4108 | spin_lock_irqsave(&n->list_lock, flags); | |
4109 | ||
4110 | list_for_each_entry(page, &n->partial, lru) { | |
434e245d | 4111 | validate_slab_slab(s, page, map); |
53e15af0 CL |
4112 | count++; |
4113 | } | |
4114 | if (count != n->nr_partial) | |
4115 | printk(KERN_ERR "SLUB %s: %ld partial slabs counted but " | |
4116 | "counter=%ld\n", s->name, count, n->nr_partial); | |
4117 | ||
4118 | if (!(s->flags & SLAB_STORE_USER)) | |
4119 | goto out; | |
4120 | ||
4121 | list_for_each_entry(page, &n->full, lru) { | |
434e245d | 4122 | validate_slab_slab(s, page, map); |
53e15af0 CL |
4123 | count++; |
4124 | } | |
4125 | if (count != atomic_long_read(&n->nr_slabs)) | |
4126 | printk(KERN_ERR "SLUB: %s %ld slabs counted but " | |
4127 | "counter=%ld\n", s->name, count, | |
4128 | atomic_long_read(&n->nr_slabs)); | |
4129 | ||
4130 | out: | |
4131 | spin_unlock_irqrestore(&n->list_lock, flags); | |
4132 | return count; | |
4133 | } | |
4134 | ||
434e245d | 4135 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
4136 | { |
4137 | int node; | |
4138 | unsigned long count = 0; | |
205ab99d | 4139 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
434e245d CL |
4140 | sizeof(unsigned long), GFP_KERNEL); |
4141 | ||
4142 | if (!map) | |
4143 | return -ENOMEM; | |
53e15af0 CL |
4144 | |
4145 | flush_all(s); | |
f64dc58c | 4146 | for_each_node_state(node, N_NORMAL_MEMORY) { |
53e15af0 CL |
4147 | struct kmem_cache_node *n = get_node(s, node); |
4148 | ||
434e245d | 4149 | count += validate_slab_node(s, n, map); |
53e15af0 | 4150 | } |
434e245d | 4151 | kfree(map); |
53e15af0 CL |
4152 | return count; |
4153 | } | |
88a420e4 | 4154 | /* |
672bba3a | 4155 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
4156 | * and freed. |
4157 | */ | |
4158 | ||
4159 | struct location { | |
4160 | unsigned long count; | |
ce71e27c | 4161 | unsigned long addr; |
45edfa58 CL |
4162 | long long sum_time; |
4163 | long min_time; | |
4164 | long max_time; | |
4165 | long min_pid; | |
4166 | long max_pid; | |
174596a0 | 4167 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 4168 | nodemask_t nodes; |
88a420e4 CL |
4169 | }; |
4170 | ||
4171 | struct loc_track { | |
4172 | unsigned long max; | |
4173 | unsigned long count; | |
4174 | struct location *loc; | |
4175 | }; | |
4176 | ||
4177 | static void free_loc_track(struct loc_track *t) | |
4178 | { | |
4179 | if (t->max) | |
4180 | free_pages((unsigned long)t->loc, | |
4181 | get_order(sizeof(struct location) * t->max)); | |
4182 | } | |
4183 | ||
68dff6a9 | 4184 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
4185 | { |
4186 | struct location *l; | |
4187 | int order; | |
4188 | ||
88a420e4 CL |
4189 | order = get_order(sizeof(struct location) * max); |
4190 | ||
68dff6a9 | 4191 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
4192 | if (!l) |
4193 | return 0; | |
4194 | ||
4195 | if (t->count) { | |
4196 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
4197 | free_loc_track(t); | |
4198 | } | |
4199 | t->max = max; | |
4200 | t->loc = l; | |
4201 | return 1; | |
4202 | } | |
4203 | ||
4204 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 4205 | const struct track *track) |
88a420e4 CL |
4206 | { |
4207 | long start, end, pos; | |
4208 | struct location *l; | |
ce71e27c | 4209 | unsigned long caddr; |
45edfa58 | 4210 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
4211 | |
4212 | start = -1; | |
4213 | end = t->count; | |
4214 | ||
4215 | for ( ; ; ) { | |
4216 | pos = start + (end - start + 1) / 2; | |
4217 | ||
4218 | /* | |
4219 | * There is nothing at "end". If we end up there | |
4220 | * we need to add something to before end. | |
4221 | */ | |
4222 | if (pos == end) | |
4223 | break; | |
4224 | ||
4225 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
4226 | if (track->addr == caddr) { |
4227 | ||
4228 | l = &t->loc[pos]; | |
4229 | l->count++; | |
4230 | if (track->when) { | |
4231 | l->sum_time += age; | |
4232 | if (age < l->min_time) | |
4233 | l->min_time = age; | |
4234 | if (age > l->max_time) | |
4235 | l->max_time = age; | |
4236 | ||
4237 | if (track->pid < l->min_pid) | |
4238 | l->min_pid = track->pid; | |
4239 | if (track->pid > l->max_pid) | |
4240 | l->max_pid = track->pid; | |
4241 | ||
174596a0 RR |
4242 | cpumask_set_cpu(track->cpu, |
4243 | to_cpumask(l->cpus)); | |
45edfa58 CL |
4244 | } |
4245 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4246 | return 1; |
4247 | } | |
4248 | ||
45edfa58 | 4249 | if (track->addr < caddr) |
88a420e4 CL |
4250 | end = pos; |
4251 | else | |
4252 | start = pos; | |
4253 | } | |
4254 | ||
4255 | /* | |
672bba3a | 4256 | * Not found. Insert new tracking element. |
88a420e4 | 4257 | */ |
68dff6a9 | 4258 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
4259 | return 0; |
4260 | ||
4261 | l = t->loc + pos; | |
4262 | if (pos < t->count) | |
4263 | memmove(l + 1, l, | |
4264 | (t->count - pos) * sizeof(struct location)); | |
4265 | t->count++; | |
4266 | l->count = 1; | |
45edfa58 CL |
4267 | l->addr = track->addr; |
4268 | l->sum_time = age; | |
4269 | l->min_time = age; | |
4270 | l->max_time = age; | |
4271 | l->min_pid = track->pid; | |
4272 | l->max_pid = track->pid; | |
174596a0 RR |
4273 | cpumask_clear(to_cpumask(l->cpus)); |
4274 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
4275 | nodes_clear(l->nodes); |
4276 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4277 | return 1; |
4278 | } | |
4279 | ||
4280 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
bbd7d57b | 4281 | struct page *page, enum track_item alloc, |
a5dd5c11 | 4282 | unsigned long *map) |
88a420e4 | 4283 | { |
a973e9dd | 4284 | void *addr = page_address(page); |
88a420e4 CL |
4285 | void *p; |
4286 | ||
39b26464 | 4287 | bitmap_zero(map, page->objects); |
5f80b13a | 4288 | get_map(s, page, map); |
88a420e4 | 4289 | |
224a88be | 4290 | for_each_object(p, s, addr, page->objects) |
45edfa58 CL |
4291 | if (!test_bit(slab_index(p, s, addr), map)) |
4292 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
4293 | } |
4294 | ||
4295 | static int list_locations(struct kmem_cache *s, char *buf, | |
4296 | enum track_item alloc) | |
4297 | { | |
e374d483 | 4298 | int len = 0; |
88a420e4 | 4299 | unsigned long i; |
68dff6a9 | 4300 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 | 4301 | int node; |
bbd7d57b ED |
4302 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
4303 | sizeof(unsigned long), GFP_KERNEL); | |
88a420e4 | 4304 | |
bbd7d57b ED |
4305 | if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
4306 | GFP_TEMPORARY)) { | |
4307 | kfree(map); | |
68dff6a9 | 4308 | return sprintf(buf, "Out of memory\n"); |
bbd7d57b | 4309 | } |
88a420e4 CL |
4310 | /* Push back cpu slabs */ |
4311 | flush_all(s); | |
4312 | ||
f64dc58c | 4313 | for_each_node_state(node, N_NORMAL_MEMORY) { |
88a420e4 CL |
4314 | struct kmem_cache_node *n = get_node(s, node); |
4315 | unsigned long flags; | |
4316 | struct page *page; | |
4317 | ||
9e86943b | 4318 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
4319 | continue; |
4320 | ||
4321 | spin_lock_irqsave(&n->list_lock, flags); | |
4322 | list_for_each_entry(page, &n->partial, lru) | |
bbd7d57b | 4323 | process_slab(&t, s, page, alloc, map); |
88a420e4 | 4324 | list_for_each_entry(page, &n->full, lru) |
bbd7d57b | 4325 | process_slab(&t, s, page, alloc, map); |
88a420e4 CL |
4326 | spin_unlock_irqrestore(&n->list_lock, flags); |
4327 | } | |
4328 | ||
4329 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 4330 | struct location *l = &t.loc[i]; |
88a420e4 | 4331 | |
9c246247 | 4332 | if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100) |
88a420e4 | 4333 | break; |
e374d483 | 4334 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
4335 | |
4336 | if (l->addr) | |
62c70bce | 4337 | len += sprintf(buf + len, "%pS", (void *)l->addr); |
88a420e4 | 4338 | else |
e374d483 | 4339 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
4340 | |
4341 | if (l->sum_time != l->min_time) { | |
e374d483 | 4342 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
f8bd2258 RZ |
4343 | l->min_time, |
4344 | (long)div_u64(l->sum_time, l->count), | |
4345 | l->max_time); | |
45edfa58 | 4346 | } else |
e374d483 | 4347 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
4348 | l->min_time); |
4349 | ||
4350 | if (l->min_pid != l->max_pid) | |
e374d483 | 4351 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
4352 | l->min_pid, l->max_pid); |
4353 | else | |
e374d483 | 4354 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
4355 | l->min_pid); |
4356 | ||
174596a0 RR |
4357 | if (num_online_cpus() > 1 && |
4358 | !cpumask_empty(to_cpumask(l->cpus)) && | |
e374d483 HH |
4359 | len < PAGE_SIZE - 60) { |
4360 | len += sprintf(buf + len, " cpus="); | |
4361 | len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
174596a0 | 4362 | to_cpumask(l->cpus)); |
45edfa58 CL |
4363 | } |
4364 | ||
62bc62a8 | 4365 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes) && |
e374d483 HH |
4366 | len < PAGE_SIZE - 60) { |
4367 | len += sprintf(buf + len, " nodes="); | |
4368 | len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
45edfa58 CL |
4369 | l->nodes); |
4370 | } | |
4371 | ||
e374d483 | 4372 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
4373 | } |
4374 | ||
4375 | free_loc_track(&t); | |
bbd7d57b | 4376 | kfree(map); |
88a420e4 | 4377 | if (!t.count) |
e374d483 HH |
4378 | len += sprintf(buf, "No data\n"); |
4379 | return len; | |
88a420e4 | 4380 | } |
ab4d5ed5 | 4381 | #endif |
88a420e4 | 4382 | |
a5a84755 CL |
4383 | #ifdef SLUB_RESILIENCY_TEST |
4384 | static void resiliency_test(void) | |
4385 | { | |
4386 | u8 *p; | |
4387 | ||
4388 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || SLUB_PAGE_SHIFT < 10); | |
4389 | ||
4390 | printk(KERN_ERR "SLUB resiliency testing\n"); | |
4391 | printk(KERN_ERR "-----------------------\n"); | |
4392 | printk(KERN_ERR "A. Corruption after allocation\n"); | |
4393 | ||
4394 | p = kzalloc(16, GFP_KERNEL); | |
4395 | p[16] = 0x12; | |
4396 | printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer" | |
4397 | " 0x12->0x%p\n\n", p + 16); | |
4398 | ||
4399 | validate_slab_cache(kmalloc_caches[4]); | |
4400 | ||
4401 | /* Hmmm... The next two are dangerous */ | |
4402 | p = kzalloc(32, GFP_KERNEL); | |
4403 | p[32 + sizeof(void *)] = 0x34; | |
4404 | printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab" | |
4405 | " 0x34 -> -0x%p\n", p); | |
4406 | printk(KERN_ERR | |
4407 | "If allocated object is overwritten then not detectable\n\n"); | |
4408 | ||
4409 | validate_slab_cache(kmalloc_caches[5]); | |
4410 | p = kzalloc(64, GFP_KERNEL); | |
4411 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
4412 | *p = 0x56; | |
4413 | printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", | |
4414 | p); | |
4415 | printk(KERN_ERR | |
4416 | "If allocated object is overwritten then not detectable\n\n"); | |
4417 | validate_slab_cache(kmalloc_caches[6]); | |
4418 | ||
4419 | printk(KERN_ERR "\nB. Corruption after free\n"); | |
4420 | p = kzalloc(128, GFP_KERNEL); | |
4421 | kfree(p); | |
4422 | *p = 0x78; | |
4423 | printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); | |
4424 | validate_slab_cache(kmalloc_caches[7]); | |
4425 | ||
4426 | p = kzalloc(256, GFP_KERNEL); | |
4427 | kfree(p); | |
4428 | p[50] = 0x9a; | |
4429 | printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", | |
4430 | p); | |
4431 | validate_slab_cache(kmalloc_caches[8]); | |
4432 | ||
4433 | p = kzalloc(512, GFP_KERNEL); | |
4434 | kfree(p); | |
4435 | p[512] = 0xab; | |
4436 | printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); | |
4437 | validate_slab_cache(kmalloc_caches[9]); | |
4438 | } | |
4439 | #else | |
4440 | #ifdef CONFIG_SYSFS | |
4441 | static void resiliency_test(void) {}; | |
4442 | #endif | |
4443 | #endif | |
4444 | ||
ab4d5ed5 | 4445 | #ifdef CONFIG_SYSFS |
81819f0f | 4446 | enum slab_stat_type { |
205ab99d CL |
4447 | SL_ALL, /* All slabs */ |
4448 | SL_PARTIAL, /* Only partially allocated slabs */ | |
4449 | SL_CPU, /* Only slabs used for cpu caches */ | |
4450 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
4451 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
4452 | }; |
4453 | ||
205ab99d | 4454 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
4455 | #define SO_PARTIAL (1 << SL_PARTIAL) |
4456 | #define SO_CPU (1 << SL_CPU) | |
4457 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 4458 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 4459 | |
62e5c4b4 CG |
4460 | static ssize_t show_slab_objects(struct kmem_cache *s, |
4461 | char *buf, unsigned long flags) | |
81819f0f CL |
4462 | { |
4463 | unsigned long total = 0; | |
81819f0f CL |
4464 | int node; |
4465 | int x; | |
4466 | unsigned long *nodes; | |
4467 | unsigned long *per_cpu; | |
4468 | ||
4469 | nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL); | |
62e5c4b4 CG |
4470 | if (!nodes) |
4471 | return -ENOMEM; | |
81819f0f CL |
4472 | per_cpu = nodes + nr_node_ids; |
4473 | ||
205ab99d CL |
4474 | if (flags & SO_CPU) { |
4475 | int cpu; | |
81819f0f | 4476 | |
205ab99d | 4477 | for_each_possible_cpu(cpu) { |
9dfc6e68 | 4478 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
ec3ab083 | 4479 | int node; |
49e22585 | 4480 | struct page *page; |
dfb4f096 | 4481 | |
bc6697d8 | 4482 | page = ACCESS_ONCE(c->page); |
ec3ab083 CL |
4483 | if (!page) |
4484 | continue; | |
205ab99d | 4485 | |
ec3ab083 CL |
4486 | node = page_to_nid(page); |
4487 | if (flags & SO_TOTAL) | |
4488 | x = page->objects; | |
4489 | else if (flags & SO_OBJECTS) | |
4490 | x = page->inuse; | |
4491 | else | |
4492 | x = 1; | |
49e22585 | 4493 | |
ec3ab083 CL |
4494 | total += x; |
4495 | nodes[node] += x; | |
4496 | ||
4497 | page = ACCESS_ONCE(c->partial); | |
49e22585 CL |
4498 | if (page) { |
4499 | x = page->pobjects; | |
bc6697d8 ED |
4500 | total += x; |
4501 | nodes[node] += x; | |
49e22585 | 4502 | } |
ec3ab083 | 4503 | |
bc6697d8 | 4504 | per_cpu[node]++; |
81819f0f CL |
4505 | } |
4506 | } | |
4507 | ||
04d94879 | 4508 | lock_memory_hotplug(); |
ab4d5ed5 | 4509 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d CL |
4510 | if (flags & SO_ALL) { |
4511 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
4512 | struct kmem_cache_node *n = get_node(s, node); | |
4513 | ||
4514 | if (flags & SO_TOTAL) | |
4515 | x = atomic_long_read(&n->total_objects); | |
4516 | else if (flags & SO_OBJECTS) | |
4517 | x = atomic_long_read(&n->total_objects) - | |
4518 | count_partial(n, count_free); | |
81819f0f | 4519 | |
81819f0f | 4520 | else |
205ab99d | 4521 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
4522 | total += x; |
4523 | nodes[node] += x; | |
4524 | } | |
4525 | ||
ab4d5ed5 CL |
4526 | } else |
4527 | #endif | |
4528 | if (flags & SO_PARTIAL) { | |
205ab99d CL |
4529 | for_each_node_state(node, N_NORMAL_MEMORY) { |
4530 | struct kmem_cache_node *n = get_node(s, node); | |
81819f0f | 4531 | |
205ab99d CL |
4532 | if (flags & SO_TOTAL) |
4533 | x = count_partial(n, count_total); | |
4534 | else if (flags & SO_OBJECTS) | |
4535 | x = count_partial(n, count_inuse); | |
81819f0f | 4536 | else |
205ab99d | 4537 | x = n->nr_partial; |
81819f0f CL |
4538 | total += x; |
4539 | nodes[node] += x; | |
4540 | } | |
4541 | } | |
81819f0f CL |
4542 | x = sprintf(buf, "%lu", total); |
4543 | #ifdef CONFIG_NUMA | |
f64dc58c | 4544 | for_each_node_state(node, N_NORMAL_MEMORY) |
81819f0f CL |
4545 | if (nodes[node]) |
4546 | x += sprintf(buf + x, " N%d=%lu", | |
4547 | node, nodes[node]); | |
4548 | #endif | |
04d94879 | 4549 | unlock_memory_hotplug(); |
81819f0f CL |
4550 | kfree(nodes); |
4551 | return x + sprintf(buf + x, "\n"); | |
4552 | } | |
4553 | ||
ab4d5ed5 | 4554 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
4555 | static int any_slab_objects(struct kmem_cache *s) |
4556 | { | |
4557 | int node; | |
81819f0f | 4558 | |
dfb4f096 | 4559 | for_each_online_node(node) { |
81819f0f CL |
4560 | struct kmem_cache_node *n = get_node(s, node); |
4561 | ||
dfb4f096 CL |
4562 | if (!n) |
4563 | continue; | |
4564 | ||
4ea33e2d | 4565 | if (atomic_long_read(&n->total_objects)) |
81819f0f CL |
4566 | return 1; |
4567 | } | |
4568 | return 0; | |
4569 | } | |
ab4d5ed5 | 4570 | #endif |
81819f0f CL |
4571 | |
4572 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
497888cf | 4573 | #define to_slab(n) container_of(n, struct kmem_cache, kobj) |
81819f0f CL |
4574 | |
4575 | struct slab_attribute { | |
4576 | struct attribute attr; | |
4577 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
4578 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
4579 | }; | |
4580 | ||
4581 | #define SLAB_ATTR_RO(_name) \ | |
ab067e99 VK |
4582 | static struct slab_attribute _name##_attr = \ |
4583 | __ATTR(_name, 0400, _name##_show, NULL) | |
81819f0f CL |
4584 | |
4585 | #define SLAB_ATTR(_name) \ | |
4586 | static struct slab_attribute _name##_attr = \ | |
ab067e99 | 4587 | __ATTR(_name, 0600, _name##_show, _name##_store) |
81819f0f | 4588 | |
81819f0f CL |
4589 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
4590 | { | |
4591 | return sprintf(buf, "%d\n", s->size); | |
4592 | } | |
4593 | SLAB_ATTR_RO(slab_size); | |
4594 | ||
4595 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
4596 | { | |
4597 | return sprintf(buf, "%d\n", s->align); | |
4598 | } | |
4599 | SLAB_ATTR_RO(align); | |
4600 | ||
4601 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
4602 | { | |
3b0efdfa | 4603 | return sprintf(buf, "%d\n", s->object_size); |
81819f0f CL |
4604 | } |
4605 | SLAB_ATTR_RO(object_size); | |
4606 | ||
4607 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
4608 | { | |
834f3d11 | 4609 | return sprintf(buf, "%d\n", oo_objects(s->oo)); |
81819f0f CL |
4610 | } |
4611 | SLAB_ATTR_RO(objs_per_slab); | |
4612 | ||
06b285dc CL |
4613 | static ssize_t order_store(struct kmem_cache *s, |
4614 | const char *buf, size_t length) | |
4615 | { | |
0121c619 CL |
4616 | unsigned long order; |
4617 | int err; | |
4618 | ||
4619 | err = strict_strtoul(buf, 10, &order); | |
4620 | if (err) | |
4621 | return err; | |
06b285dc CL |
4622 | |
4623 | if (order > slub_max_order || order < slub_min_order) | |
4624 | return -EINVAL; | |
4625 | ||
4626 | calculate_sizes(s, order); | |
4627 | return length; | |
4628 | } | |
4629 | ||
81819f0f CL |
4630 | static ssize_t order_show(struct kmem_cache *s, char *buf) |
4631 | { | |
834f3d11 | 4632 | return sprintf(buf, "%d\n", oo_order(s->oo)); |
81819f0f | 4633 | } |
06b285dc | 4634 | SLAB_ATTR(order); |
81819f0f | 4635 | |
73d342b1 DR |
4636 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
4637 | { | |
4638 | return sprintf(buf, "%lu\n", s->min_partial); | |
4639 | } | |
4640 | ||
4641 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
4642 | size_t length) | |
4643 | { | |
4644 | unsigned long min; | |
4645 | int err; | |
4646 | ||
4647 | err = strict_strtoul(buf, 10, &min); | |
4648 | if (err) | |
4649 | return err; | |
4650 | ||
c0bdb232 | 4651 | set_min_partial(s, min); |
73d342b1 DR |
4652 | return length; |
4653 | } | |
4654 | SLAB_ATTR(min_partial); | |
4655 | ||
49e22585 CL |
4656 | static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf) |
4657 | { | |
4658 | return sprintf(buf, "%u\n", s->cpu_partial); | |
4659 | } | |
4660 | ||
4661 | static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf, | |
4662 | size_t length) | |
4663 | { | |
4664 | unsigned long objects; | |
4665 | int err; | |
4666 | ||
4667 | err = strict_strtoul(buf, 10, &objects); | |
4668 | if (err) | |
4669 | return err; | |
74ee4ef1 DR |
4670 | if (objects && kmem_cache_debug(s)) |
4671 | return -EINVAL; | |
49e22585 CL |
4672 | |
4673 | s->cpu_partial = objects; | |
4674 | flush_all(s); | |
4675 | return length; | |
4676 | } | |
4677 | SLAB_ATTR(cpu_partial); | |
4678 | ||
81819f0f CL |
4679 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
4680 | { | |
62c70bce JP |
4681 | if (!s->ctor) |
4682 | return 0; | |
4683 | return sprintf(buf, "%pS\n", s->ctor); | |
81819f0f CL |
4684 | } |
4685 | SLAB_ATTR_RO(ctor); | |
4686 | ||
81819f0f CL |
4687 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
4688 | { | |
4689 | return sprintf(buf, "%d\n", s->refcount - 1); | |
4690 | } | |
4691 | SLAB_ATTR_RO(aliases); | |
4692 | ||
81819f0f CL |
4693 | static ssize_t partial_show(struct kmem_cache *s, char *buf) |
4694 | { | |
d9acf4b7 | 4695 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
4696 | } |
4697 | SLAB_ATTR_RO(partial); | |
4698 | ||
4699 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
4700 | { | |
d9acf4b7 | 4701 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
4702 | } |
4703 | SLAB_ATTR_RO(cpu_slabs); | |
4704 | ||
4705 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
4706 | { | |
205ab99d | 4707 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
4708 | } |
4709 | SLAB_ATTR_RO(objects); | |
4710 | ||
205ab99d CL |
4711 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
4712 | { | |
4713 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
4714 | } | |
4715 | SLAB_ATTR_RO(objects_partial); | |
4716 | ||
49e22585 CL |
4717 | static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf) |
4718 | { | |
4719 | int objects = 0; | |
4720 | int pages = 0; | |
4721 | int cpu; | |
4722 | int len; | |
4723 | ||
4724 | for_each_online_cpu(cpu) { | |
4725 | struct page *page = per_cpu_ptr(s->cpu_slab, cpu)->partial; | |
4726 | ||
4727 | if (page) { | |
4728 | pages += page->pages; | |
4729 | objects += page->pobjects; | |
4730 | } | |
4731 | } | |
4732 | ||
4733 | len = sprintf(buf, "%d(%d)", objects, pages); | |
4734 | ||
4735 | #ifdef CONFIG_SMP | |
4736 | for_each_online_cpu(cpu) { | |
4737 | struct page *page = per_cpu_ptr(s->cpu_slab, cpu) ->partial; | |
4738 | ||
4739 | if (page && len < PAGE_SIZE - 20) | |
4740 | len += sprintf(buf + len, " C%d=%d(%d)", cpu, | |
4741 | page->pobjects, page->pages); | |
4742 | } | |
4743 | #endif | |
4744 | return len + sprintf(buf + len, "\n"); | |
4745 | } | |
4746 | SLAB_ATTR_RO(slabs_cpu_partial); | |
4747 | ||
a5a84755 CL |
4748 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) |
4749 | { | |
4750 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
4751 | } | |
4752 | ||
4753 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
4754 | const char *buf, size_t length) | |
4755 | { | |
4756 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
4757 | if (buf[0] == '1') | |
4758 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
4759 | return length; | |
4760 | } | |
4761 | SLAB_ATTR(reclaim_account); | |
4762 | ||
4763 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
4764 | { | |
4765 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); | |
4766 | } | |
4767 | SLAB_ATTR_RO(hwcache_align); | |
4768 | ||
4769 | #ifdef CONFIG_ZONE_DMA | |
4770 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
4771 | { | |
4772 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
4773 | } | |
4774 | SLAB_ATTR_RO(cache_dma); | |
4775 | #endif | |
4776 | ||
4777 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
4778 | { | |
4779 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU)); | |
4780 | } | |
4781 | SLAB_ATTR_RO(destroy_by_rcu); | |
4782 | ||
ab9a0f19 LJ |
4783 | static ssize_t reserved_show(struct kmem_cache *s, char *buf) |
4784 | { | |
4785 | return sprintf(buf, "%d\n", s->reserved); | |
4786 | } | |
4787 | SLAB_ATTR_RO(reserved); | |
4788 | ||
ab4d5ed5 | 4789 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
4790 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) |
4791 | { | |
4792 | return show_slab_objects(s, buf, SO_ALL); | |
4793 | } | |
4794 | SLAB_ATTR_RO(slabs); | |
4795 | ||
205ab99d CL |
4796 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) |
4797 | { | |
4798 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
4799 | } | |
4800 | SLAB_ATTR_RO(total_objects); | |
4801 | ||
81819f0f CL |
4802 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
4803 | { | |
4804 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE)); | |
4805 | } | |
4806 | ||
4807 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
4808 | const char *buf, size_t length) | |
4809 | { | |
4810 | s->flags &= ~SLAB_DEBUG_FREE; | |
b789ef51 CL |
4811 | if (buf[0] == '1') { |
4812 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4813 | s->flags |= SLAB_DEBUG_FREE; |
b789ef51 | 4814 | } |
81819f0f CL |
4815 | return length; |
4816 | } | |
4817 | SLAB_ATTR(sanity_checks); | |
4818 | ||
4819 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
4820 | { | |
4821 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
4822 | } | |
4823 | ||
4824 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
4825 | size_t length) | |
4826 | { | |
4827 | s->flags &= ~SLAB_TRACE; | |
b789ef51 CL |
4828 | if (buf[0] == '1') { |
4829 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4830 | s->flags |= SLAB_TRACE; |
b789ef51 | 4831 | } |
81819f0f CL |
4832 | return length; |
4833 | } | |
4834 | SLAB_ATTR(trace); | |
4835 | ||
81819f0f CL |
4836 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) |
4837 | { | |
4838 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
4839 | } | |
4840 | ||
4841 | static ssize_t red_zone_store(struct kmem_cache *s, | |
4842 | const char *buf, size_t length) | |
4843 | { | |
4844 | if (any_slab_objects(s)) | |
4845 | return -EBUSY; | |
4846 | ||
4847 | s->flags &= ~SLAB_RED_ZONE; | |
b789ef51 CL |
4848 | if (buf[0] == '1') { |
4849 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4850 | s->flags |= SLAB_RED_ZONE; |
b789ef51 | 4851 | } |
06b285dc | 4852 | calculate_sizes(s, -1); |
81819f0f CL |
4853 | return length; |
4854 | } | |
4855 | SLAB_ATTR(red_zone); | |
4856 | ||
4857 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
4858 | { | |
4859 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
4860 | } | |
4861 | ||
4862 | static ssize_t poison_store(struct kmem_cache *s, | |
4863 | const char *buf, size_t length) | |
4864 | { | |
4865 | if (any_slab_objects(s)) | |
4866 | return -EBUSY; | |
4867 | ||
4868 | s->flags &= ~SLAB_POISON; | |
b789ef51 CL |
4869 | if (buf[0] == '1') { |
4870 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4871 | s->flags |= SLAB_POISON; |
b789ef51 | 4872 | } |
06b285dc | 4873 | calculate_sizes(s, -1); |
81819f0f CL |
4874 | return length; |
4875 | } | |
4876 | SLAB_ATTR(poison); | |
4877 | ||
4878 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
4879 | { | |
4880 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
4881 | } | |
4882 | ||
4883 | static ssize_t store_user_store(struct kmem_cache *s, | |
4884 | const char *buf, size_t length) | |
4885 | { | |
4886 | if (any_slab_objects(s)) | |
4887 | return -EBUSY; | |
4888 | ||
4889 | s->flags &= ~SLAB_STORE_USER; | |
b789ef51 CL |
4890 | if (buf[0] == '1') { |
4891 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4892 | s->flags |= SLAB_STORE_USER; |
b789ef51 | 4893 | } |
06b285dc | 4894 | calculate_sizes(s, -1); |
81819f0f CL |
4895 | return length; |
4896 | } | |
4897 | SLAB_ATTR(store_user); | |
4898 | ||
53e15af0 CL |
4899 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
4900 | { | |
4901 | return 0; | |
4902 | } | |
4903 | ||
4904 | static ssize_t validate_store(struct kmem_cache *s, | |
4905 | const char *buf, size_t length) | |
4906 | { | |
434e245d CL |
4907 | int ret = -EINVAL; |
4908 | ||
4909 | if (buf[0] == '1') { | |
4910 | ret = validate_slab_cache(s); | |
4911 | if (ret >= 0) | |
4912 | ret = length; | |
4913 | } | |
4914 | return ret; | |
53e15af0 CL |
4915 | } |
4916 | SLAB_ATTR(validate); | |
a5a84755 CL |
4917 | |
4918 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) | |
4919 | { | |
4920 | if (!(s->flags & SLAB_STORE_USER)) | |
4921 | return -ENOSYS; | |
4922 | return list_locations(s, buf, TRACK_ALLOC); | |
4923 | } | |
4924 | SLAB_ATTR_RO(alloc_calls); | |
4925 | ||
4926 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
4927 | { | |
4928 | if (!(s->flags & SLAB_STORE_USER)) | |
4929 | return -ENOSYS; | |
4930 | return list_locations(s, buf, TRACK_FREE); | |
4931 | } | |
4932 | SLAB_ATTR_RO(free_calls); | |
4933 | #endif /* CONFIG_SLUB_DEBUG */ | |
4934 | ||
4935 | #ifdef CONFIG_FAILSLAB | |
4936 | static ssize_t failslab_show(struct kmem_cache *s, char *buf) | |
4937 | { | |
4938 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB)); | |
4939 | } | |
4940 | ||
4941 | static ssize_t failslab_store(struct kmem_cache *s, const char *buf, | |
4942 | size_t length) | |
4943 | { | |
4944 | s->flags &= ~SLAB_FAILSLAB; | |
4945 | if (buf[0] == '1') | |
4946 | s->flags |= SLAB_FAILSLAB; | |
4947 | return length; | |
4948 | } | |
4949 | SLAB_ATTR(failslab); | |
ab4d5ed5 | 4950 | #endif |
53e15af0 | 4951 | |
2086d26a CL |
4952 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
4953 | { | |
4954 | return 0; | |
4955 | } | |
4956 | ||
4957 | static ssize_t shrink_store(struct kmem_cache *s, | |
4958 | const char *buf, size_t length) | |
4959 | { | |
4960 | if (buf[0] == '1') { | |
4961 | int rc = kmem_cache_shrink(s); | |
4962 | ||
4963 | if (rc) | |
4964 | return rc; | |
4965 | } else | |
4966 | return -EINVAL; | |
4967 | return length; | |
4968 | } | |
4969 | SLAB_ATTR(shrink); | |
4970 | ||
81819f0f | 4971 | #ifdef CONFIG_NUMA |
9824601e | 4972 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 4973 | { |
9824601e | 4974 | return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
4975 | } |
4976 | ||
9824601e | 4977 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
4978 | const char *buf, size_t length) |
4979 | { | |
0121c619 CL |
4980 | unsigned long ratio; |
4981 | int err; | |
4982 | ||
4983 | err = strict_strtoul(buf, 10, &ratio); | |
4984 | if (err) | |
4985 | return err; | |
4986 | ||
e2cb96b7 | 4987 | if (ratio <= 100) |
0121c619 | 4988 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 4989 | |
81819f0f CL |
4990 | return length; |
4991 | } | |
9824601e | 4992 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
4993 | #endif |
4994 | ||
8ff12cfc | 4995 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
4996 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
4997 | { | |
4998 | unsigned long sum = 0; | |
4999 | int cpu; | |
5000 | int len; | |
5001 | int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL); | |
5002 | ||
5003 | if (!data) | |
5004 | return -ENOMEM; | |
5005 | ||
5006 | for_each_online_cpu(cpu) { | |
9dfc6e68 | 5007 | unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si]; |
8ff12cfc CL |
5008 | |
5009 | data[cpu] = x; | |
5010 | sum += x; | |
5011 | } | |
5012 | ||
5013 | len = sprintf(buf, "%lu", sum); | |
5014 | ||
50ef37b9 | 5015 | #ifdef CONFIG_SMP |
8ff12cfc CL |
5016 | for_each_online_cpu(cpu) { |
5017 | if (data[cpu] && len < PAGE_SIZE - 20) | |
50ef37b9 | 5018 | len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]); |
8ff12cfc | 5019 | } |
50ef37b9 | 5020 | #endif |
8ff12cfc CL |
5021 | kfree(data); |
5022 | return len + sprintf(buf + len, "\n"); | |
5023 | } | |
5024 | ||
78eb00cc DR |
5025 | static void clear_stat(struct kmem_cache *s, enum stat_item si) |
5026 | { | |
5027 | int cpu; | |
5028 | ||
5029 | for_each_online_cpu(cpu) | |
9dfc6e68 | 5030 | per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0; |
78eb00cc DR |
5031 | } |
5032 | ||
8ff12cfc CL |
5033 | #define STAT_ATTR(si, text) \ |
5034 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
5035 | { \ | |
5036 | return show_stat(s, buf, si); \ | |
5037 | } \ | |
78eb00cc DR |
5038 | static ssize_t text##_store(struct kmem_cache *s, \ |
5039 | const char *buf, size_t length) \ | |
5040 | { \ | |
5041 | if (buf[0] != '0') \ | |
5042 | return -EINVAL; \ | |
5043 | clear_stat(s, si); \ | |
5044 | return length; \ | |
5045 | } \ | |
5046 | SLAB_ATTR(text); \ | |
8ff12cfc CL |
5047 | |
5048 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
5049 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
5050 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
5051 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
5052 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
5053 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
5054 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
5055 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
5056 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
5057 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
e36a2652 | 5058 | STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch); |
8ff12cfc CL |
5059 | STAT_ATTR(FREE_SLAB, free_slab); |
5060 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
5061 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
5062 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
5063 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
5064 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
5065 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
03e404af | 5066 | STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass); |
65c3376a | 5067 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
b789ef51 CL |
5068 | STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail); |
5069 | STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail); | |
49e22585 CL |
5070 | STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc); |
5071 | STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free); | |
8028dcea AS |
5072 | STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node); |
5073 | STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain); | |
8ff12cfc CL |
5074 | #endif |
5075 | ||
06428780 | 5076 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
5077 | &slab_size_attr.attr, |
5078 | &object_size_attr.attr, | |
5079 | &objs_per_slab_attr.attr, | |
5080 | &order_attr.attr, | |
73d342b1 | 5081 | &min_partial_attr.attr, |
49e22585 | 5082 | &cpu_partial_attr.attr, |
81819f0f | 5083 | &objects_attr.attr, |
205ab99d | 5084 | &objects_partial_attr.attr, |
81819f0f CL |
5085 | &partial_attr.attr, |
5086 | &cpu_slabs_attr.attr, | |
5087 | &ctor_attr.attr, | |
81819f0f CL |
5088 | &aliases_attr.attr, |
5089 | &align_attr.attr, | |
81819f0f CL |
5090 | &hwcache_align_attr.attr, |
5091 | &reclaim_account_attr.attr, | |
5092 | &destroy_by_rcu_attr.attr, | |
a5a84755 | 5093 | &shrink_attr.attr, |
ab9a0f19 | 5094 | &reserved_attr.attr, |
49e22585 | 5095 | &slabs_cpu_partial_attr.attr, |
ab4d5ed5 | 5096 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
5097 | &total_objects_attr.attr, |
5098 | &slabs_attr.attr, | |
5099 | &sanity_checks_attr.attr, | |
5100 | &trace_attr.attr, | |
81819f0f CL |
5101 | &red_zone_attr.attr, |
5102 | &poison_attr.attr, | |
5103 | &store_user_attr.attr, | |
53e15af0 | 5104 | &validate_attr.attr, |
88a420e4 CL |
5105 | &alloc_calls_attr.attr, |
5106 | &free_calls_attr.attr, | |
ab4d5ed5 | 5107 | #endif |
81819f0f CL |
5108 | #ifdef CONFIG_ZONE_DMA |
5109 | &cache_dma_attr.attr, | |
5110 | #endif | |
5111 | #ifdef CONFIG_NUMA | |
9824601e | 5112 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
5113 | #endif |
5114 | #ifdef CONFIG_SLUB_STATS | |
5115 | &alloc_fastpath_attr.attr, | |
5116 | &alloc_slowpath_attr.attr, | |
5117 | &free_fastpath_attr.attr, | |
5118 | &free_slowpath_attr.attr, | |
5119 | &free_frozen_attr.attr, | |
5120 | &free_add_partial_attr.attr, | |
5121 | &free_remove_partial_attr.attr, | |
5122 | &alloc_from_partial_attr.attr, | |
5123 | &alloc_slab_attr.attr, | |
5124 | &alloc_refill_attr.attr, | |
e36a2652 | 5125 | &alloc_node_mismatch_attr.attr, |
8ff12cfc CL |
5126 | &free_slab_attr.attr, |
5127 | &cpuslab_flush_attr.attr, | |
5128 | &deactivate_full_attr.attr, | |
5129 | &deactivate_empty_attr.attr, | |
5130 | &deactivate_to_head_attr.attr, | |
5131 | &deactivate_to_tail_attr.attr, | |
5132 | &deactivate_remote_frees_attr.attr, | |
03e404af | 5133 | &deactivate_bypass_attr.attr, |
65c3376a | 5134 | &order_fallback_attr.attr, |
b789ef51 CL |
5135 | &cmpxchg_double_fail_attr.attr, |
5136 | &cmpxchg_double_cpu_fail_attr.attr, | |
49e22585 CL |
5137 | &cpu_partial_alloc_attr.attr, |
5138 | &cpu_partial_free_attr.attr, | |
8028dcea AS |
5139 | &cpu_partial_node_attr.attr, |
5140 | &cpu_partial_drain_attr.attr, | |
81819f0f | 5141 | #endif |
4c13dd3b DM |
5142 | #ifdef CONFIG_FAILSLAB |
5143 | &failslab_attr.attr, | |
5144 | #endif | |
5145 | ||
81819f0f CL |
5146 | NULL |
5147 | }; | |
5148 | ||
5149 | static struct attribute_group slab_attr_group = { | |
5150 | .attrs = slab_attrs, | |
5151 | }; | |
5152 | ||
5153 | static ssize_t slab_attr_show(struct kobject *kobj, | |
5154 | struct attribute *attr, | |
5155 | char *buf) | |
5156 | { | |
5157 | struct slab_attribute *attribute; | |
5158 | struct kmem_cache *s; | |
5159 | int err; | |
5160 | ||
5161 | attribute = to_slab_attr(attr); | |
5162 | s = to_slab(kobj); | |
5163 | ||
5164 | if (!attribute->show) | |
5165 | return -EIO; | |
5166 | ||
5167 | err = attribute->show(s, buf); | |
5168 | ||
5169 | return err; | |
5170 | } | |
5171 | ||
5172 | static ssize_t slab_attr_store(struct kobject *kobj, | |
5173 | struct attribute *attr, | |
5174 | const char *buf, size_t len) | |
5175 | { | |
5176 | struct slab_attribute *attribute; | |
5177 | struct kmem_cache *s; | |
5178 | int err; | |
5179 | ||
5180 | attribute = to_slab_attr(attr); | |
5181 | s = to_slab(kobj); | |
5182 | ||
5183 | if (!attribute->store) | |
5184 | return -EIO; | |
5185 | ||
5186 | err = attribute->store(s, buf, len); | |
5187 | ||
5188 | return err; | |
5189 | } | |
5190 | ||
151c602f CL |
5191 | static void kmem_cache_release(struct kobject *kobj) |
5192 | { | |
5193 | struct kmem_cache *s = to_slab(kobj); | |
5194 | ||
84c1cf62 | 5195 | kfree(s->name); |
151c602f CL |
5196 | kfree(s); |
5197 | } | |
5198 | ||
52cf25d0 | 5199 | static const struct sysfs_ops slab_sysfs_ops = { |
81819f0f CL |
5200 | .show = slab_attr_show, |
5201 | .store = slab_attr_store, | |
5202 | }; | |
5203 | ||
5204 | static struct kobj_type slab_ktype = { | |
5205 | .sysfs_ops = &slab_sysfs_ops, | |
151c602f | 5206 | .release = kmem_cache_release |
81819f0f CL |
5207 | }; |
5208 | ||
5209 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
5210 | { | |
5211 | struct kobj_type *ktype = get_ktype(kobj); | |
5212 | ||
5213 | if (ktype == &slab_ktype) | |
5214 | return 1; | |
5215 | return 0; | |
5216 | } | |
5217 | ||
9cd43611 | 5218 | static const struct kset_uevent_ops slab_uevent_ops = { |
81819f0f CL |
5219 | .filter = uevent_filter, |
5220 | }; | |
5221 | ||
27c3a314 | 5222 | static struct kset *slab_kset; |
81819f0f CL |
5223 | |
5224 | #define ID_STR_LENGTH 64 | |
5225 | ||
5226 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
5227 | * |
5228 | * Format :[flags-]size | |
81819f0f CL |
5229 | */ |
5230 | static char *create_unique_id(struct kmem_cache *s) | |
5231 | { | |
5232 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
5233 | char *p = name; | |
5234 | ||
5235 | BUG_ON(!name); | |
5236 | ||
5237 | *p++ = ':'; | |
5238 | /* | |
5239 | * First flags affecting slabcache operations. We will only | |
5240 | * get here for aliasable slabs so we do not need to support | |
5241 | * too many flags. The flags here must cover all flags that | |
5242 | * are matched during merging to guarantee that the id is | |
5243 | * unique. | |
5244 | */ | |
5245 | if (s->flags & SLAB_CACHE_DMA) | |
5246 | *p++ = 'd'; | |
5247 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
5248 | *p++ = 'a'; | |
5249 | if (s->flags & SLAB_DEBUG_FREE) | |
5250 | *p++ = 'F'; | |
5a896d9e VN |
5251 | if (!(s->flags & SLAB_NOTRACK)) |
5252 | *p++ = 't'; | |
81819f0f CL |
5253 | if (p != name + 1) |
5254 | *p++ = '-'; | |
5255 | p += sprintf(p, "%07d", s->size); | |
5256 | BUG_ON(p > name + ID_STR_LENGTH - 1); | |
5257 | return name; | |
5258 | } | |
5259 | ||
5260 | static int sysfs_slab_add(struct kmem_cache *s) | |
5261 | { | |
5262 | int err; | |
5263 | const char *name; | |
5264 | int unmergeable; | |
5265 | ||
97d06609 | 5266 | if (slab_state < FULL) |
81819f0f CL |
5267 | /* Defer until later */ |
5268 | return 0; | |
5269 | ||
5270 | unmergeable = slab_unmergeable(s); | |
5271 | if (unmergeable) { | |
5272 | /* | |
5273 | * Slabcache can never be merged so we can use the name proper. | |
5274 | * This is typically the case for debug situations. In that | |
5275 | * case we can catch duplicate names easily. | |
5276 | */ | |
27c3a314 | 5277 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
5278 | name = s->name; |
5279 | } else { | |
5280 | /* | |
5281 | * Create a unique name for the slab as a target | |
5282 | * for the symlinks. | |
5283 | */ | |
5284 | name = create_unique_id(s); | |
5285 | } | |
5286 | ||
27c3a314 | 5287 | s->kobj.kset = slab_kset; |
1eada11c GKH |
5288 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name); |
5289 | if (err) { | |
5290 | kobject_put(&s->kobj); | |
81819f0f | 5291 | return err; |
1eada11c | 5292 | } |
81819f0f CL |
5293 | |
5294 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
5788d8ad XF |
5295 | if (err) { |
5296 | kobject_del(&s->kobj); | |
5297 | kobject_put(&s->kobj); | |
81819f0f | 5298 | return err; |
5788d8ad | 5299 | } |
81819f0f CL |
5300 | kobject_uevent(&s->kobj, KOBJ_ADD); |
5301 | if (!unmergeable) { | |
5302 | /* Setup first alias */ | |
5303 | sysfs_slab_alias(s, s->name); | |
5304 | kfree(name); | |
5305 | } | |
5306 | return 0; | |
5307 | } | |
5308 | ||
5309 | static void sysfs_slab_remove(struct kmem_cache *s) | |
5310 | { | |
97d06609 | 5311 | if (slab_state < FULL) |
2bce6485 CL |
5312 | /* |
5313 | * Sysfs has not been setup yet so no need to remove the | |
5314 | * cache from sysfs. | |
5315 | */ | |
5316 | return; | |
5317 | ||
81819f0f CL |
5318 | kobject_uevent(&s->kobj, KOBJ_REMOVE); |
5319 | kobject_del(&s->kobj); | |
151c602f | 5320 | kobject_put(&s->kobj); |
81819f0f CL |
5321 | } |
5322 | ||
5323 | /* | |
5324 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 5325 | * available lest we lose that information. |
81819f0f CL |
5326 | */ |
5327 | struct saved_alias { | |
5328 | struct kmem_cache *s; | |
5329 | const char *name; | |
5330 | struct saved_alias *next; | |
5331 | }; | |
5332 | ||
5af328a5 | 5333 | static struct saved_alias *alias_list; |
81819f0f CL |
5334 | |
5335 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
5336 | { | |
5337 | struct saved_alias *al; | |
5338 | ||
97d06609 | 5339 | if (slab_state == FULL) { |
81819f0f CL |
5340 | /* |
5341 | * If we have a leftover link then remove it. | |
5342 | */ | |
27c3a314 GKH |
5343 | sysfs_remove_link(&slab_kset->kobj, name); |
5344 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
5345 | } |
5346 | ||
5347 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
5348 | if (!al) | |
5349 | return -ENOMEM; | |
5350 | ||
5351 | al->s = s; | |
5352 | al->name = name; | |
5353 | al->next = alias_list; | |
5354 | alias_list = al; | |
5355 | return 0; | |
5356 | } | |
5357 | ||
5358 | static int __init slab_sysfs_init(void) | |
5359 | { | |
5b95a4ac | 5360 | struct kmem_cache *s; |
81819f0f CL |
5361 | int err; |
5362 | ||
2bce6485 CL |
5363 | down_write(&slub_lock); |
5364 | ||
0ff21e46 | 5365 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); |
27c3a314 | 5366 | if (!slab_kset) { |
2bce6485 | 5367 | up_write(&slub_lock); |
81819f0f CL |
5368 | printk(KERN_ERR "Cannot register slab subsystem.\n"); |
5369 | return -ENOSYS; | |
5370 | } | |
5371 | ||
97d06609 | 5372 | slab_state = FULL; |
26a7bd03 | 5373 | |
5b95a4ac | 5374 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 5375 | err = sysfs_slab_add(s); |
5d540fb7 CL |
5376 | if (err) |
5377 | printk(KERN_ERR "SLUB: Unable to add boot slab %s" | |
5378 | " to sysfs\n", s->name); | |
26a7bd03 | 5379 | } |
81819f0f CL |
5380 | |
5381 | while (alias_list) { | |
5382 | struct saved_alias *al = alias_list; | |
5383 | ||
5384 | alias_list = alias_list->next; | |
5385 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 CL |
5386 | if (err) |
5387 | printk(KERN_ERR "SLUB: Unable to add boot slab alias" | |
068ce415 | 5388 | " %s to sysfs\n", al->name); |
81819f0f CL |
5389 | kfree(al); |
5390 | } | |
5391 | ||
2bce6485 | 5392 | up_write(&slub_lock); |
81819f0f CL |
5393 | resiliency_test(); |
5394 | return 0; | |
5395 | } | |
5396 | ||
5397 | __initcall(slab_sysfs_init); | |
ab4d5ed5 | 5398 | #endif /* CONFIG_SYSFS */ |
57ed3eda PE |
5399 | |
5400 | /* | |
5401 | * The /proc/slabinfo ABI | |
5402 | */ | |
158a9624 | 5403 | #ifdef CONFIG_SLABINFO |
57ed3eda PE |
5404 | static void print_slabinfo_header(struct seq_file *m) |
5405 | { | |
5406 | seq_puts(m, "slabinfo - version: 2.1\n"); | |
3b0efdfa | 5407 | seq_puts(m, "# name <active_objs> <num_objs> <object_size> " |
57ed3eda PE |
5408 | "<objperslab> <pagesperslab>"); |
5409 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); | |
5410 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); | |
5411 | seq_putc(m, '\n'); | |
5412 | } | |
5413 | ||
5414 | static void *s_start(struct seq_file *m, loff_t *pos) | |
5415 | { | |
5416 | loff_t n = *pos; | |
5417 | ||
5418 | down_read(&slub_lock); | |
5419 | if (!n) | |
5420 | print_slabinfo_header(m); | |
5421 | ||
5422 | return seq_list_start(&slab_caches, *pos); | |
5423 | } | |
5424 | ||
5425 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
5426 | { | |
5427 | return seq_list_next(p, &slab_caches, pos); | |
5428 | } | |
5429 | ||
5430 | static void s_stop(struct seq_file *m, void *p) | |
5431 | { | |
5432 | up_read(&slub_lock); | |
5433 | } | |
5434 | ||
5435 | static int s_show(struct seq_file *m, void *p) | |
5436 | { | |
5437 | unsigned long nr_partials = 0; | |
5438 | unsigned long nr_slabs = 0; | |
5439 | unsigned long nr_inuse = 0; | |
205ab99d CL |
5440 | unsigned long nr_objs = 0; |
5441 | unsigned long nr_free = 0; | |
57ed3eda PE |
5442 | struct kmem_cache *s; |
5443 | int node; | |
5444 | ||
5445 | s = list_entry(p, struct kmem_cache, list); | |
5446 | ||
5447 | for_each_online_node(node) { | |
5448 | struct kmem_cache_node *n = get_node(s, node); | |
5449 | ||
5450 | if (!n) | |
5451 | continue; | |
5452 | ||
5453 | nr_partials += n->nr_partial; | |
5454 | nr_slabs += atomic_long_read(&n->nr_slabs); | |
205ab99d CL |
5455 | nr_objs += atomic_long_read(&n->total_objects); |
5456 | nr_free += count_partial(n, count_free); | |
57ed3eda PE |
5457 | } |
5458 | ||
205ab99d | 5459 | nr_inuse = nr_objs - nr_free; |
57ed3eda PE |
5460 | |
5461 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse, | |
834f3d11 CL |
5462 | nr_objs, s->size, oo_objects(s->oo), |
5463 | (1 << oo_order(s->oo))); | |
57ed3eda PE |
5464 | seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0); |
5465 | seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs, | |
5466 | 0UL); | |
5467 | seq_putc(m, '\n'); | |
5468 | return 0; | |
5469 | } | |
5470 | ||
7b3c3a50 | 5471 | static const struct seq_operations slabinfo_op = { |
57ed3eda PE |
5472 | .start = s_start, |
5473 | .next = s_next, | |
5474 | .stop = s_stop, | |
5475 | .show = s_show, | |
5476 | }; | |
5477 | ||
7b3c3a50 AD |
5478 | static int slabinfo_open(struct inode *inode, struct file *file) |
5479 | { | |
5480 | return seq_open(file, &slabinfo_op); | |
5481 | } | |
5482 | ||
5483 | static const struct file_operations proc_slabinfo_operations = { | |
5484 | .open = slabinfo_open, | |
5485 | .read = seq_read, | |
5486 | .llseek = seq_lseek, | |
5487 | .release = seq_release, | |
5488 | }; | |
5489 | ||
5490 | static int __init slab_proc_init(void) | |
5491 | { | |
ab067e99 | 5492 | proc_create("slabinfo", S_IRUSR, NULL, &proc_slabinfo_operations); |
7b3c3a50 AD |
5493 | return 0; |
5494 | } | |
5495 | module_init(slab_proc_init); | |
158a9624 | 5496 | #endif /* CONFIG_SLABINFO */ |